Car Audio Explained, Part 10: Time Alignment, Levels, Listening Tests, Processors and Common Tuning Mistakes (Andy Wehmeyer's Lessons)

This is Part 10 of 19 in the series Andy Wehmeyer on Car Audio, Explained for Beginners. Start here · Download the full guide as a PDF

Unofficial guide. Andy Wehmeyer's points are paraphrased from his public forum posts, blog, tuning guides, videos and podcasts, with a link to each source; the plain-English explanations were written for this guide. Not written, reviewed or endorsed by Andy Wehmeyer or Audiofrog.

3.9 Time alignment and delay

What this part is about: Time alignment means adding tiny electronic delays so that sound from speakers at different distances reaches your ears at the same moment. This part covers what delay can and cannot do, how he measures and enters it, and the many situations where he says not to use it.

3.9 #1 · "Delay doesn't move speakers."

It only changes what happens between them; it can't move the left/right edges or push the stage forward or back (a friend leaving 30 minutes later doesn't move their house). Test: play the left channel alone and add delay; the image stays put. A DSP can only add delay, so you delay the nearer speakers to match the farthest. [RARE] Sources: Audiofrog blog Dec 2020, Time Alignment Part 1 · DIYMA 2021-03-26

In plain English: In a car you sit much closer to the speakers on your side, so their sound reaches you first. A DSP (digital signal processor, the small computer that shapes the sound before the amplifiers) can hold a speaker's signal back by a tiny amount; this is called delay or time alignment. Many people believe delay "moves" a speaker, pushing it farther away or pushing the stage forward. His point is that it doesn't. Delay only changes how the sound of two speakers combines in the space between them, for example where a centered voice appears. The outer edges of the stage stay at the speakers. His analogy: a friend who leaves home 30 minutes later hasn't moved their house. You can test it yourself: play only the left speaker and add delay; the sound stays exactly where it was. And because a DSP can only hold sound back, never speed it up, you delay the nearer speakers to match the farthest one.

3.9 #2 · What delay is for: crossovers and the phantom center both assume equal arrival.

A 27–28" left/right path difference (typical door mids) puts the first cancellation near 250 Hz, mid-vocal, repeating at odd multiples; delaying the near side about 2 ms (27 ÷ 13,582 in/s) fixes it at that seat. Delay also puts sub and fronts in phase: "nobody hates coherent bass." Sources: Audiofrog blog Dec 2020, Time Alignment Part 2 · Tuning PDF v2 (~2023) · OFCAP Ep. 38 @1:00:03

In plain English: Two things in a sound system only work if the sounds arrive together. A crossover splits the music between speakers (bass to the woofer, treble to the tweeter) and counts on the two halves arriving in step so they blend. The "phantom center" is the illusion of a singer floating between the left and right speakers, and it needs equal arrival too. With speakers in the doors, the far one is typically 27–28 inches (about 70 cm) farther from you. That gap means that near 250 Hz, the middle of a singing voice, one speaker's wave is pushing while the other's is pulling, so they cancel; the same cancellation repeats at 3, 5, 7 times that frequency. Sound travels about 13,582 inches per second, so 27 inches takes about 2 ms (two thousandths of a second). Delaying the near side by that much fixes it for that seat. Delay also lines up the subwoofer and front speakers so the bass works together: in his words, "nobody hates coherent bass."

3.9 #3 · How to measure and enter delays.

Put tape on the headrest center (or the mic tip) and measure to the center of each grille or dust cap. Key the front delays to the midrange, never an average of mid and tweeter ("just makes both wrong"). Subtract each distance from the largest (sub 70", tweeter 32" → tweeter gets 38", sub 0); ms = inches ÷ 13,584 × 1000. Know whether your DSP wants distance, time, or the subtracted value. Sources: Tuning PDF v1 (~2018) · DIYMA 2022-02-22 · DIYMA 2021-09-26

In plain English: This is his measuring procedure, step by step. First, put a piece of tape at the center of the headrest (or use the microphone tip) as the reference point for your head. Second, measure with a tape measure from there to the center of each speaker's grille or dust cap (the dome in the middle of a cone). Third, for the front speakers use the midrange's distance, not the tweeter's and not an average of the two; he says averaging "just makes both wrong." In the source post he gives his reasons: we hear phase errors more at lower frequencies, and the mids' delays help place the center image. Fourth, subtract each distance from the largest one. In his example the sub is 70 inches away and the tweeter 32 inches, so the tweeter gets 38 inches of delay and the sub, being farthest, gets none. Fifth, to turn inches into milliseconds, divide by 13,584 (the speed of sound in inches per second) and multiply by 1000; 38 inches is about 2.8 ms. Finally, check your DSP's manual: some want the raw distance, some a time, and some the subtracted value. Typing in the wrong kind is a common mistake.

3.9 #4 · A tape measure is accurate enough; extra precision is wasted.

At 80 Hz 1" ≈ 2° and 10" ≈ 20°; at 60 Hz a foot of error is only ~19°. A 1" error caused nothing in his midbass example, and LR4 forgives listeners of different heights. DSP resolution (one sample at 48 kHz ≈ 0.28") puts its cancellation at 24 kHz, so 1 mm resolution is "nonsense." Set delays before crossovers, then leave them alone. [RARE] Sources: Tuning PDF v1 (~2018) · DIYMA 2021-03-26 · DIYMA 2021-09-02

In plain English: Many people believe time alignment must be set to the millimeter, using a microphone and software. His point is that a tape measure is close enough. Background: "phase" describes where a sound wave is in its up-and-down cycle, measured in degrees; 360° is one full cycle, and 180° means exactly opposite, which cancels. Bass waves are long, so a small distance error barely shifts them: at 80 Hz an inch is about 2° and ten inches about 20°; at 60 Hz a whole foot of error is only about 19°. (He treats differences under 90° as hard to hear; see 3.8 #6.) In his midbass example a 1-inch error caused nothing, and an LR4 crossover (Linkwitz-Riley, 24 dB per octave, the type he recommends) forgives listeners of different heights. A DSP's smallest delay step at a 48 kHz sample rate is one sample, about 0.28 inch, and an error that size would only cancel at 24 kHz, above the range of human hearing, so 1 mm settings are "nonsense." So in practice: set delays before crossovers, then leave them alone.

3.9 #5 · Impulse peaks aren't arrival times for low-passed drivers, so impulse alignment rarely beats a tape.

The peak's slope depends on high-frequency content a sub or midbass doesn't play (REW's own help says so); the true arrival is the zero crossing before the right peak. If you must: crossovers off (midbass low-pass off), back up from the first tall peak to −12 dB, take the initial not the tallest peak. A $100 DSSF3 cross-correlation method costs hours and lands no closer. [RARE] Sources: DIYMA 2013 · DIYMA 2021-03-26 · OFCAP Ep. 19 @3:25:01

In plain English: An "impulse response" is a measurement of what the mic picks up after a speaker plays a very short click. Software such as REW (Room EQ Wizard, a free measurement program) draws it as a spiky graph, and many people read the tallest spike as the moment the sound arrived. His point is that for speakers that play only low notes (a sub, or a midbass with a low-pass filter that removes its highs), the peak is not the arrival time. How steep the peak is depends on high frequencies those speakers don't play; REW's own help says so. The true arrival is where the trace crosses zero just before the correct peak. If you must use impulses: switch the crossovers off (including the midbass low-pass), find the first tall peak, back up from it to the point 12 dB lower, and use the first peak, not the tallest. He says a cross-correlation method using DSSF3 (a $100 measurement program) costs hours and lands no closer than a tape measure.

3.9 #6 · Automatic time alignment mis-measures dash speakers.

It picks impulse peaks shaped by the highs, and a dash speaker's highs reach the mic via the windshield, so it measures speaker → glass → mic. Equal L/R errors are harmless; unequal ones cancel where the error is half a wavelength (a 2 kHz dip ≈ 3.5" ≈ 0.26 ms). [RARE] Sources: DIYMA 2022-03-24

In plain English: Some DSPs and measurement programs set delays automatically from a mic measurement. They look for peaks in the impulse response, and those peaks are shaped mostly by the high frequencies. A speaker on top of the dash usually fires up at the windshield, so its highs reach the mic mainly by bouncing off the glass. The software then measures the longer path, speaker to glass to mic, as if it were the speaker's distance. If the error is the same on the left and the right, no harm is done. If it differs, the two sides cancel at the frequency where the difference equals half a wavelength (half the length of one sound wave). His example: a dip at 2 kHz matches an error of about 3.5 inches (9 cm), or 0.26 ms.

3.9 #7 · The invert-and-find-the-null test rarely works.

It needs even-order crossovers, closely matched levels and both drivers EQ'd to the target slopes an octave each side, and reflections can still spoil it; it suits 12 dB Butterworth, not LR4. Sources: DIYMA 2021-09-02 · DIYMA 2015

In plain English: A popular trick for setting delay between two speakers that share a crossover works like this: reverse one speaker's polarity (its electrical plus and minus), play sound around the crossover frequency, adjust delay until the sound is quietest (the "null"), then switch the polarity back. He says this rarely works. It needs even-order crossovers (filter slopes such as 12 or 24 dB per octave), closely matched levels, and both speakers already equalized to the intended filter shape for an octave (a doubling or halving of frequency) on each side of the crossover. Even then, reflections in the car can spoil the result. It suits a 12 dB per octave Butterworth crossover, not the LR4 type he recommends. In the source thread his advice is to compensate for distance with a tape measure instead.

3.9 #8 · Never set delays by ear or use them to move the image or add "depth."

Ear-setting needs matched level and response that usually don't exist yet; delay tweaks move the image but "screw up" other frequencies. He rejects score-sheet advice to "play with delays" for stage depth. Of TrackRite-style tools, only the distance part works. [RARE] Sources: Tuning PDF v2 (~2023) · DIYMA 2018-05-22 · OFCAP Ep. 19 @3:23:07

In plain English: Many people believe the way to tune is to sit in the car and nudge delays by ear until the singer sounds right, or to use delays to make the stage seem deeper. He says never. Setting delay by ear only works if left and right are already matched in level and frequency response, and at that stage they usually aren't. Changing a delay can move the image, but it "screws up" other frequencies. He rejects the advice on competition score sheets (the judges' forms) to "play with delays" for stage depth. TrackRite is a commercial tuning aid; of tools like it, he says only the part that sets delays from distance works, not the parts meant for crossovers or for moving the image.

3.9 #9 · Delay is for distance only, never for crossover phase or group delay.

Music is mostly steady state, so there is no need to make up the cycle of lag across a 24 dB crossover. Car left and right speakers differ, so re-tweaking mid delays for crossover phase likely ruins the center ("a haircut using a cartoon image"). Use distance delays, LR4 and EQ each driver to that acoustic target. [RARE] Sources: DIYMA 2015 · DIYMA 2021-03-26

In plain English: Background: a crossover filter doesn't only divide frequencies; it also makes some of them lag slightly behind others. This lag is called group delay, and across a 24 dB per octave crossover it adds up to about one full cycle. Some tuners add extra delay to make up for it. He says delay is for distance only. Music is mostly "steady state" (sustained sound rather than isolated clicks), so there's no need to make up that lag. And because a car's left and right speakers sit in different spots and behave differently, re-tweaking the midrange delays to fix crossover phase will likely ruin the center image. His full comparison is giving yourself a haircut using a cartoon of you drawn by someone else: you are working from a picture that doesn't match reality. So in practice: set delays by distance, use LR4 crossovers, and EQ each speaker until its measured response matches that LR4 shape.

3.9 #10 · Reflections, cabin gain, corner loading and resonances don't change delay settings.

As long as a direct path exists, the direct sound arrives first and sets the delay; a reflection matters only if it is louder than the direct sound (only possible for a tweeter aimed away). Reflections make EQ useful, not different delays. The one judgement call: a narrow spike followed by a bigger, wider one may be an adjacent reflection, so find the surface first. [RARE] Sources: DIYMA 2021-03-26 · DIYMA 2013

In plain English: Background: "cabin gain" is the bass boost a small closed space like a car gives; "corner loading" is the boost a speaker gets near surfaces; resonances are frequencies where parts of the car ring. Many people think these call for different delays. He says they don't. If a speaker has a clear straight line to your ears, its direct sound always arrives first, and that sets the delay. A reflection (a bounce off glass or a panel) only matters if it is louder than the direct sound, which can happen only with a tweeter aimed away from you. Reflections are a reason to use EQ, not to change delays. The one judgement call: on an impulse graph, a narrow spike followed by a bigger, wider one may be the direct sound plus a reflection from a nearby surface. In the source post he says to check with a tape measure and find that surface before deciding which spike is the real arrival.

3.9 #11 · Align drivers within one side too; only their overlap band combs.

At 1 kHz (13.6" wavelength) a mid 6.8" farther than the tweeter is 180° out; 0.5 ms on the tweeter fixes it. Once a cancellation is pushed above audibility or outside the pair's passband it doesn't matter, so small midbass errors below 200 Hz are inaudible. Aim for everything to seem to come from one point above the dash, no "rainbow." Sources: Audiofrog blog Dec 2020, Time Alignment Part 3 · DIYMA 2020-10-03

In plain English: Delay isn't only for left versus right. On the same side, a midrange and a tweeter at different distances can also interfere, but only in the band where both play (the overlap around their crossover). That interference is called comb filtering: a row of evenly spaced dips. His example: at 1 kHz one wave is 13.6 inches long, so if the midrange is 6.8 inches (half a wave) farther away than the tweeter, the two are 180° apart and cancel; 0.5 ms of delay on the tweeter fixes it. Once a cancellation is pushed above what we can hear, or outside the range the pair plays together, it no longer matters, so small errors between midbass speakers below 200 Hz are inaudible. His goal: everything should seem to come from one point above the dash, with no "rainbow." The sources don't define "rainbow" exactly; it appears to mean different frequency ranges seeming to come from different places.

3.9 #12 · Don't time-align the fronts to the sub.

Harman's repeated trials found no audible or measurable difference in bass placement below a 100 Hz crossover; an 80 Hz wave is ~14 ft, so a sub must be ~7 ft farther to be 180° off (a polarity swap fixes that). If bass seems behind you, check 60–160 Hz for dips, lower the sub, or cross sub 60 Hz/midbass 80 Hz, since raising sub level moves the acoustic crossover; last resort, delay all fronts equally. [RARE] [Changed view: he argued "over and over" for sub alignment at Harman, then lost.] [Conflict: the Tuning PDFs tape-measure the sub like any other speaker.] Sources: DIYMA 2013 · DIYMA 2021-03-26 · Audiofrog blog Dec 2020, Time Alignment Part 3

In plain English: Harman is the large audio company (owner of JBL and other brands) where he worked. A sub in the trunk is much farther away than the front speakers, so many people delay the fronts to line up with it. He says don't. Harman's repeated tests found no audible or measurable difference in where the bass seemed to come from, with the sub crossed below 100 Hz. Bass waves are long: an 80 Hz wave is about 14 feet (4.3 m), so the sub would need to be about 7 feet farther away to be fully opposite (180°), and a polarity swap would fix that. If the bass seems to come from behind you, he suggests checking 60–160 Hz for dips, turning the sub down, or crossing the sub at 60 Hz and the midbass at 80 Hz, because turning the sub up shifts where the two actually blend (the acoustic crossover). Last resort: delay all front speakers by the same amount. The tags note that he argued for sub alignment at Harman "over and over," then lost; and that his Tuning PDFs measure the sub with a tape like any other speaker, which conflicts with this advice.

3.9 #13 · Factory delay and DSP latency both sneak into your timing.

Most factory systems with outboard amps already contain time alignment, and almost no "un-EQ" integration DSP removes it; the leftover delay combs and makes your settings wrong. Run every amp through the processor: MS-8's ~8 ms latency means a sub fed straight from the head unit arrives early, and total latency can cause hands-free echo. [RARE] Sources: DIYMA 2013 · DIYMA MS-8 thread · DIYMA (MS-8 replacement)

In plain English: Factory systems with a separate amplifier often contain their own processing. "Integration" DSPs take the factory signal and try to "un-EQ" it, undoing the factory sound shaping so you start clean. He says most factory systems with outboard amps already include time alignment, and almost none of these DSPs remove it. The leftover factory delay causes comb filtering and makes your own settings wrong. Second, every processor takes a little time to work, called latency; the MS-8 (a JBL processor from his Harman years) takes about 8 ms. If one amp, such as the sub's, is fed straight from the head unit (the radio) while the others go through the processor, the sub's sound arrives early. So in practice: run every amp through the processor. Total latency can also cause an echo on hands-free phone calls.

3.9 #14 · Delay is a one-seat tool: it doubles the other seat's error.

With ~100 cm vs ~150 cm paths, delaying the near side ~1.5 ms doubles the passenger's error. For two seats use placement (kick panels or any spot that keeps the path difference under about a foot), all-pass phase EQ (3.8 #14–15) or a center channel. [Changed view: "forget the passenger" is a cop-out; a good passenger seat no longer costs the driver's seat.] Sources: DIYMA 2016 · DIYMA 2021-06-29 · OFCAP Ep. 38 @52:04

In plain English: Delay fixes the timing for one seat only. Say the driver is about 100 cm (39 inches) from the left speaker and 150 cm (59 inches) from the right. Delaying the left side by about 1.5 ms fixes the driver's seat. But the passenger already hears the right speaker first, and the delay makes the left one arrive even later for them, so their error doubles. For good sound in both seats he points to speaker placement (kick panels, the panels in the footwells, or any spot that keeps the left/right distance difference under about a foot), all-pass "phase EQ" (see 3.8 #14–15), or a center channel. The tag marks a change of view: he now calls "forget the passenger" a cop-out, because a good passenger seat no longer has to cost the driver's seat anything.

3.9 #15 · With a center channel and upmixer, drop the left/right delays.

The center carries the center image, and center-to-side path differences are small, pushing phase problems up where they matter less; EQ left, center and right alike and image with level only. Toole's no-delay-with-a-center advice works only if the center covers all the midbass the mains play. [RARE] Sources: SQOLOGY Part 2 @14:02 · DIYMA 2015

In plain English: A center channel is a speaker in the middle of the dash. An upmixer is processing (such as Logic 7 or Dolby Pro Logic II) that pulls centered sounds out of a stereo recording and sends them to that speaker. He says that with this setup you remove the left/right delays. The center speaker now carries the center image itself, so delay is no longer needed to create one. The distance differences between the center and each side speaker are small, which pushes any phase problems up to higher frequencies where they matter less. EQ the left, center and right speakers to sound alike, and place the image with level only. Floyd Toole, a well-known loudspeaker researcher, advises no delay with a center; he says that works only if the center plays all the midbass the left and right speakers play.

3.9 #16 · A wide ~6 dB dip in the L+R sum can be fixed by trimming the driver-side delay while you watch it.

First make sure it isn't one mic spot (move the mic; check single readings). A 2nd-order all-pass suits narrow, deep dips; for a wide one, reduce the driver's-side delay with both sides playing and keep the change if nothing else breaks. [RARE] Sources: DIYMA 2022-03-24

In plain English: "L+R sum" means measuring the left and right sides playing together. A dip of about 6 dB (a clear drop in loudness) spread over a wide range of frequencies suggests the two sides are partly out of step there. First make sure it's real: if you measured at one mic spot, move the mic a little; if you used an average, take several single readings around the area. If the problem shows up in only one spot, ignore it. A 2nd-order all-pass filter (a phase-shifting filter, see 3.8 #12) suits narrow, deep dips. For a wide one, he says: with both sides playing, reduce the driver's-side delay while watching the dip on screen, and if it improves without creating another problem, keep the change.

3.9 #17 · To "de-tune" for listeners who dislike a center image, move delays deliberately; never switch them off.

Active 3-way: delay the midbass but not mids and highs. 2-way plus sub: shorten the near-side delay so drivers stay in phase near the ~100 Hz crossover but not above, or flip mids and tweeters while keeping midbass in phase. Save the correct tune as a second preset. Correct is the start of expertise; backing off it to please a listener is the rest. [RARE] Sources: OFCAP Ep. 38 @1:02:06 · CMA Connected 2021 @42:16

In plain English: Some listeners dislike a tightly centered image (see 3.10 #7). For them he deliberately loosens the tune instead of switching delays off; he says switching them off costs bass coherence. He explains that in phase you hear an image between the speakers, and out of phase you hear the speakers themselves. In an active 3-way (separate amp channels for midbass, midrange and tweeter): delay the midbass, but not the midrange and tweeter. In a 2-way plus sub: shorten the near-side delay so the speakers stay in step around the roughly 100 Hz crossover but not above it, or flip the polarity of the mids and tweeters while keeping the midbass in phase. Save the correct tune as a second preset so you can switch back. His view: getting it correct is the start of expertise; knowing how to back off from correct to please a listener is the rest.

3.9 #18 · Rear fill: plain in-phase stereo, turned up until it interferes, then delayed 10–20 ms.

High-pass the rears at ≥100–200 Hz; add delay until they stop sounding like separate sources (more level needs more delay; ~35 ms becomes an echo; 1 ms ≈ 1 ft); in small cars set level and delay per side. [RARE] [Changed view: in 2017 he still gave recipes for differential rear fill (L−R, delayed 12–20 ms); by late 2021 he said he'd helped popularize it but now finds it nasal, phasey and hard to tune.] Sources: DIYMA 2017-10-06 (old view) · DIYMA 2021-12-02 · OFCAP Ep. 19 @3:33:00

In plain English: "Rear fill" means playing some sound through the rear speakers for a sense of space while the image stays in front. His later (2021) recipe: feed the rears ordinary stereo, in phase. Tune the front first, then turn the rears up until they start to interfere with the front image, and stop. High-pass them (remove their bass) at 100–200 Hz or higher. Add delay until they no longer sound like separate sources and the image is back in front; 10–20 ms (hundredths of a second) is about right, and more level needs more delay. At around 35 ms the delay becomes a distinct echo. Sound travels roughly 1 foot per millisecond. In small cars, set level and delay for each side separately. The tag: in 2017 he still gave recipes for "differential" rear fill (the difference signal, left minus right, delayed 12–20 ms); by late 2021 he said he had helped popularize it but now finds it nasal, phasey and hard to tune.

3.9 #19 · Measure an L−R (difference) feed one side at a time, and treat center leakage as a clue.

Mono noise into L−R gives silence, so play one channel or pan the balance fully (L − 0 = L). Center content leaking into an L−R output means something upstream isn't matched: noise not truly mono, unequal input EQ or levels, head-unit phase or delay differences, or lossy noise files. Sources: DIYMA 2021-05-11 · DIYMA 2021-12-02

In plain English: An L−R (left minus right) signal contains only what differs between the two channels; anything identical in both cancels out. So if you measure such an output with mono pink noise (the same hiss-like test signal in both channels), you get silence. Instead, play just one channel, or turn the balance fully to one side: with nothing on the right, L minus 0 is simply L. If centered sound leaks into an L−R output, he says something earlier in the chain isn't matched: the noise isn't truly mono, the input EQ or levels differ between left and right, the head unit (radio) adds different phase or delay to each side, or the noise files are lossy (compressed, like an MP3).

3.9 #20 · Stage "width" from a delay error is just out-of-phase combing.

An extra 1 ms on one side starts cancellations at 500 Hz, then 1.5 kHz, 2.5 kHz and so on; in those bands you hear the two speakers, which passes for width, and in between they fuse into an image. [RARE] Sources: DIYMA 2017-10-01

In plain English: Some people like the "wide" stage they hear when delays are off or wrong. He says that width is just an error. If one side has an extra 1 ms (one thousandth of a second) of delay, the two sides cancel at 500 Hz, then again at 1.5 kHz, 2.5 kHz and so on, a pattern called comb filtering. In those bands you hear the two speakers as separate sources, which passes for width; in the bands between, they fuse into a single image. So what sounds like extra width is really this out-of-phase pattern.

3.9 #21 · How judged cars game the score, and why that isn't the goal for customers.

Competitors optimize per judging track; without per-track EQ he'd skip delay and use an all-pass for the center-vocal track. Everything above the vocal fundamental (about 100–250, maybe 300 Hz) can be out of phase without moving the vocal, and judges hear those out-of-phase highs as stage width. Winning isn't great audio; for customers, get close to correct. [RARE] Sources: OFCAP Ep. 38 @54:01

In plain English: In sound-quality competitions, judges score cars on specific test tracks, and competitors tune for each judging track. He says that if he couldn't switch EQ between tracks, he would skip delay and use an all-pass filter (a phase-shifting filter) for the centered-vocal track. A voice's "fundamental" is its lowest, main pitch, here about 100–250 Hz, maybe 300 Hz. Everything above it can be out of phase without moving the vocal from the center, and judges hear those out-of-phase highs, where the separate speakers become audible, as a wide stage. He adds that he has sat in cars scored for a "wide stage" that had no center image on ordinary music. His conclusion: winning isn't the same as great audio; for customers, get close to correct.

3.10 Level matching and centering the image

What this part is about: After the delays, the next job is making the left and right sides equally loud at every frequency, so that a centered singer really sounds centered. This part covers how he matches levels, how he moves the center image, and why a centered image is a goal not every listener shares.

3.10 #1 · Delay is the easy half; matching left and right level and response is the hard half.

Time alignment fixes only the phase half; in one member's car the channels differed by almost 20 dB at 80 Hz, which no delay can fix. Center-panned sound is mono, so each ear needs identical level from both sides at every frequency; "4 or 5 bands of parametric EQ aren't enough." Match overall L/R within "a couple of dB" with channel levels before EQ. [RARE] Sources: DIYMA 2008 · DIYMA 2017-09-12 · Tuning PDF v1 (~2018)

In plain English: A centered image needs two things: the left and right sounds must arrive in step (phase, which delay handles) and be equally loud. "Frequency response" is how loud a speaker plays at each pitch, from deep bass to high treble. He says delay is the easy half. The hard half is making left and right match in level and response. In one forum member's car the two sides differed by almost 20 dB at 80 Hz, a huge difference that no delay can fix. A singer panned to the center is mono, meaning the same signal in both channels, so each ear needs the same level from both sides at every frequency. That takes a lot of EQ: "4 or 5 bands of parametric EQ aren't enough." (A parametric EQ band is one adjustable boost or cut at a chosen frequency.) So in practice: first get the overall left/right levels within "a couple of dB" using the channel level controls, then EQ.

3.10 #2 · Center the image with level, never delay: move the whole left or right front.

If the center sits left, turn the entire left side down a little (and vice versa); ignore what the level controls end up reading. Off-center in a left-hand-drive car, expect the right side to need more level. Level, unlike delay, can't be set by measurement alone: the mic lumps direct sound and reflections together, while we hear the direct sound as location. Used last, it has worked in "hundreds of cars." [RARE] Sources: Tuning PDF v2 (~2023) · DIYMA 2022-03-19 · DIYMA 2022-03-01

In plain English: Once delays and EQ are done, if a centered singer still leans to one side, he moves it with level, never delay. If the center sits to the left, turn the whole left front side (all its speakers together) down a little; if it sits right, turn the right side down. Ignore what the level controls read afterward. In a left-hand-drive car, where you sit to the left, expect the right side to need more level. Some argue that needing to lower one side means the delays are wrong; he disagrees. Delay can be set by measurement, but level can't: a microphone hears the direct sound and all its reflections as one sound, while our ears take the location from the direct sound. So a meter can read "equal" when your ears disagree. He says doing this last has worked in "hundreds of cars."

3.10 #3 · Diagnose a missing or off-center image before touching levels.

A smeared or one-sided singer usually means delays entered wrong (e.g., distances typed into a DSP that wants delays), a reversed mid or midbass, or a left/right response mismatch; recheck delays with the tape, then the L/R match, and only then level. Sources: Absolute Electronix livestream @43:32 · DIYMA 2017-10-31 · DIYMA 2018-05-22

In plain English: If the singer sounds smeared across the dash or stuck on one side, don't reach for the level controls first. He says the usual causes are: delays entered wrong (for example, typing measured distances into a DSP that expects delay values), a midrange or midbass wired with reversed polarity (plus and minus swapped), or left and right not matching in frequency response. His order: re-check the delays with the tape measure, then check the left/right match, and only then adjust level.

3.10 #4 · To center on the driver instead of the dash, turn the far side down about 6 dB.

Include the sub if it has two inputs; he says this moves the image about halfway between dash center and the left speaker. With arrival and level matched, the center lands midway between the speakers, not in front of the driver, and a 6 dB difference moves it halfway toward the louder side. A near speaker at 18" vs a far one at 36" is already about 6 dB louder, which feels like too much when set by hand. [RARE] Sources: Tuning PDF v1 (~2018) · DIYMA 2017-07-07 · DIYMA 2012

In plain English: With delays set and levels matched, a centered singer appears midway between the two speakers, in the middle of the dash, not straight in front of the driver. Some people want it in front of the steering wheel instead. He says to turn the far side down about 6 dB (a 6 dB cut is a quarter of the power), including the sub if it takes two inputs (left and right). A 6 dB difference moves the image halfway toward the louder side, so he says it lands about halfway between the dash center and the left speaker. Distance alone also changes level: a speaker 18 inches away is about 6 dB louder at your ear than one 36 inches away. In a 2012 post he says matching those takes about a 6 dB cut on the near side, which can feel like too much when you set it by hand.

3.10 #5 · The phantom image is vector math: balance only steers where the speakers are in phase.

Without delay correction the summed vector leans to the nearer speaker; with it, the simple level-based sum returns. Even with correct delay, a 6 dB L/R difference over half an octave pulls that band toward the louder side, and level-based localization works well into the midrange, not just highs. [RARE] Sources: DIYMA 2015 · DIYMA 2015 · DIYMA 2018-05-22

In plain English: A "vector" is a quantity with both a size and a direction, like an arrow. His point is that the phantom image comes from adding the left and right sounds like arrows, which takes both their level and their timing into account. With delays uncorrected, the combined result leans toward the nearer speaker. With delays corrected, the simple rule returns: the louder side pulls the image toward it, which is what the balance control relies on. Even with correct delays, if one side is 6 dB louder over just half an octave (a narrow slice of frequencies), that slice is pulled toward the louder side. A common belief is that level differences only steer high frequencies; he says level-based localization works well into the midrange too.

3.10 #6 · Last step, noise and gain: move every amp gain by the same amount.

If hiss on the silent track bothers you, lower all amplifier input-sensitivity controls equally; for more volume raise them equally until hiss appears, then back off. In the attributed gain guide, amp gains are set acoustically against the target (DSP channels at −12 dB), not with a scope, tweeter amp down first; the sub is raised to target with mono noise, and if only one input feeds the sub expect to add about 6 dB. Sources: Tuning PDF v2 (~2023) · DIYMA gain guide (attributed)

In plain English: An amplifier's gain (input-sensitivity) control sets how strongly the amp responds to the incoming signal. It isn't a volume knob, but it affects how loud the system can play and how much background hiss you hear. His final step: play the silent track on the test disc. If the hiss bothers you, lower every amp's gain by the same amount, so the balance you tuned doesn't change. If you want more volume, raise them all equally until hiss appears, then back off. A forum guide titled "Andy's guide to setting gains" (credited to him only by its title) sets amp gains acoustically, by measuring the sound in the car against the target curve, not with an oscilloscope (an instrument that displays voltage). In it, the DSP channels sit at −12 dB and the tweeter amp is turned down first; the sub is raised to the target using mono noise, and if only one input feeds the sub, expect to add about 6 dB.

3.10 #7 · A centered image is a preference, not a universal goal.

Two famous record producers heard his demo car: one loved the imaging, the other heard the center as "mono" and hated it. People who prefer processing off "aren't used to hearing true stereo" ("audiophile mode" vs "amusement park mode"); give them a de-tuned preset (3.9 #17). Sources: CMA Connected 2021 @41:19 · DIYMA 2011

In plain English: Not everyone wants a sharp, centered singer. Two famous record producers listened to his demo car: one loved the imaging; the other heard the center as "mono" and hated it. He says people who prefer the processing switched off "aren't used to hearing true stereo." He joked about an "audiophile mode" and an "amusement park mode." So in practice: give such listeners a de-tuned preset, as described in 3.9 #17.

3.10 #8 · Many cars image well quietly and poorly loud.

Cabin reflections spread images, more so at high volume. Sources: DIYMA 2010 · DIYMA 2022-02-23

In plain English: Reflections are sound bouncing off the glass and panels before it reaches you. He says they spread images out, more so at high volume. So a car can image well at low volume and poorly when turned up.

3.10 #9 · He tolerates frequency-response errors more than spatial ones.

A peak or dip bothers him less than a missing center image. Sources: SQOLOGY Part 2 @46:07

In plain English: A frequency-response error is a tonal one: some pitches too loud (a peak) or too quiet (a dip). A spatial error is about where sounds appear. He says a peak or dip bothers him less than a missing center image, so he puts up with tonal errors more readily than spatial ones.

3.11 Listening checks and test tracks

What this part is about: Measuring sets up the tune; listening checks it. This part covers when he listens, what sounds he uses (band-limited pink noise, panned bass guitar and piano rather than sine tones or singers), and what a good result should and shouldn't be expected to do.

3.11 #1 · Listening verifies the tune; it doesn't set it up.

He never listens until the measurements say the car is ready, and doesn't iterate by ear between steps ("you don't taste raw chicken after salting it"). Early listening misleads because small level differences are hard to hear and delay or response errors shift the image. If hours of touch-up follow the objective steps, the measurement, the tuner's grasp of it, or the process is wrong. Sources: DIYMA 2017-06-28 · UMI-1 video @24:03 · DIYMA 2021-07-09

In plain English: Many tuners adjust a little, listen, and adjust again, over and over. He doesn't. He measures and sets everything first, and only listens once the measurements say the car is ready; listening is the check at the end, not the method. He doesn't go back and forth by ear between steps. His analogy: you don't taste raw chicken after salting it. Listening early misleads you, because small level differences are hard to hear, and delay or response errors shift the image around. And if a tune still needs hours of touch-up by ear after the measured steps, something is wrong: the measurement, the tuner's understanding of it, or the process itself.

3.11 #2 · The first reference a tuner needs is the sound of two speakers in phase, not "a concert."

Many tuners have never sat between two speakers and flipped one to hear the difference. Training method: buy any matched $5 thrift-store pair, set them in an equilateral triangle, sit dead center and play music; imaging is just what two similar speakers do for a centered listener. [RARE] Sources: Absolute Electronix livestream @41:56 · Absolute Electronix livestream @46:15

In plain English: Many people believe a tuner's goal is to make a car sound like a live concert. He says the first reference you need is simpler: the sound of two speakers in phase. Many tuners have never sat between two speakers and flipped one (reversed its polarity) to hear the difference. His training method: buy any matched pair of speakers, even a $5 pair from a thrift store, set them up so they and you form an equilateral triangle (all three sides equal), sit dead center and play music. That teaches you what imaging is: simply what two similar speakers do for a listener sitting in the middle.

3.11 #3 · Check placement band by band with mono band-limited pink noise; each band should sit centered.

In the 2021 video: track 2 (20–160 Hz) must come from the dash center in front of you, track 3 (80–400 Hz) checks midbass, then 400 Hz–1 kHz and 400 Hz–20 kHz. Highs spread more, which is acceptable (dash and windshield reflections). For a quick center check, mono pink noise band-passed 100 Hz–3 kHz. [Conflict: the Tuning PDFs number these tracks 2–5.] Sources: UMI-1 video @27:01 · Tuning PDF v2 (~2023) · DIYMA 2011

In plain English: Pink noise is a hiss-like test sound containing all frequencies. "Band-limited" means filtered so only one range plays, and "mono" means identical in both channels, so it should appear dead center. The tracks referred to are on the test disc that goes with his tuning process. In his 2021 video: track 2 (20–160 Hz, the bass) must come from the dash center in front of you; track 3 (80–400 Hz) checks the midbass; further tracks cover 400 Hz–1 kHz and 400 Hz–20 kHz. Each band should sit centered. Higher frequencies spread out more, which is acceptable because of reflections from the dash and windshield. For a quick center check, use mono pink noise band-passed (filtered) to 100 Hz–3 kHz. The tag flags that the Tuning PDFs number these tracks 2–5, differently from the video.

3.11 #4 · Use panning tracks of bass guitar and piano, not singers or "seven drum beats."

Singers aren't always mixed center, and snare tracks have no midbass. The CD pans pink noise (tracks 6–10), bass guitar (11–15, checks midbass delay and level) and piano (16–20, should spread wider) across five positions; half-left should land exactly halfway between center and far left. Track 25 is for the last-resort front delay, track 38 (silence) for noise. [RARE] [Conflict: the attributed gain guide still suggests "7 drum beats" for centering.] Sources: UMI-1 video @28:00 · Tuning PDF v1 (~2018) · DIYMA 2013

In plain English: Many people check their stage with a singer, or with a track of seven drum beats panned from left to right. He says singers aren't always mixed in the center, and snare drum tracks contain no midbass. Instead, his test CD pans three sounds across five positions from left to right: pink noise (tracks 6–10), bass guitar (tracks 11–15, which check the midbass delay and level) and piano (tracks 16–20, which should spread wider). The half-left track should land exactly halfway between the center and far left. Track 25 is for the last-resort front delay (see 3.9 #12), and track 38, which is silence, is for checking noise (see 3.10 #6). The tag notes a conflict: the gain guide attributed to him still suggests "7 drum beats" for centering.

3.11 #5 · Listen with band-limited noise or music, never pure sine tones.

Above 1 kHz the reflection comb is so dense that a tone easily lands in a null, and the nulls differ left to right; that is why 1980s tone-based centering (and "31-sine" tuning) fails, while music and head movement erase the comb. If you must tune by ear, use band-limited pink noise. [RARE] Sources: DIYMA 2017-07-12 · DIYMA 2021-07-08

In plain English: A sine tone is a single pure pitch, like a tuning fork. In a car, reflections create a pattern of dips (nulls) where sound cancels, packed very tightly above 1 kHz. A single tone can easily land in one of those nulls, and the nulls fall in different places for the left and right speakers, so a tone misleads you. That is why the 1980s method of centering with tones, and "31-sine" tuning, fail. Music covers many frequencies at once, and we naturally move our heads, which erases the comb of dips. If you must tune by ear, use band-limited pink noise (hiss filtered to one range).

3.11 #6 · Don't tune for depth or height; the stage comes from the recording and the car.

Images fall inside the triangle of you and the speakers; depth and width come from the recording's ambience ("there's no trick" beyond it), and cabin reflections only spread images. A static ~3.5 kHz "height" bump works only while you're adjusting it, then it's just a response error. [RARE] Sources: Tuning PDF v2 (~2023) · DIYMA 2021-11-11 · UMI-1 video @31:05

In plain English: Many tuners try to make the stage seem deeper (farther away) or higher (up on the windshield) with EQ or delay tricks. He says not to tune for that. Images appear inside the triangle formed by you and the two speakers. Depth and width come from the ambience captured in the recording (the sense of the space it was recorded in), and "there's no trick" beyond that; reflections in the cabin only spread images out. A popular height trick is a fixed boost around 3.5 kHz. He says it works only while you're adjusting it; after that it's just an error in the frequency response.

3.11 #7 · Missing midbass "snap" usually lives around 1–3.5 kHz, not in the midbass.

Look near 1 kHz first; flat dynamics are often the mastering (a 6 dB crest factor). If door-midbass impact is truly missing, raise the midbass/sub crossover and work the 60–160 Hz EQ. [RARE] Sources: DIYMA 2015 · DIYMA 2011

In plain English: "Snap" or "punch" is the sharp, impactful attack of a kick drum or bass note. When it's missing, many people turn up the midbass. He says the missing snap usually lives around 1–3.5 kHz, not in the midbass, so look near 1 kHz first. Flat, lifeless dynamics are often down to the recording's mastering. "Crest factor" is the gap between a recording's loudest peaks and its average level; a crest factor of only 6 dB means heavily squashed music. If the impact from the door midbass speakers really is missing, raise the midbass/sub crossover frequency and work on the EQ between 60 and 160 Hz.

3.11 #8 · How experienced listeners judge a car in seconds.

Full, impactful bass that seems to come from the front; a centered singer; then width and depth, which partly belong to the recording. Sources: Absolute Electronix livestream @53:21

In plain English: He describes what experienced listeners check first when they sit in a car: bass that is full and impactful and seems to come from the front (not the trunk); a singer who sits in the center; and then the width and depth of the stage, which partly depend on the recording rather than the car.

3.11 #9 · Listen to factory premium systems for what they do well.

Aftermarket listeners judge them on bass and loudness and miss the envelopment and two-seat imaging OEM engineers built in. [RARE] Sources: Absolute Electronix livestream @56:27

In plain English: Factory premium systems are the branded sound systems carmakers offer as upgrades. OEM (original equipment manufacturer) means the carmaker and its suppliers; aftermarket means equipment installed later. He says aftermarket enthusiasts judge factory systems on bass and loudness, and so miss what OEM engineers built in: envelopment (the feeling of being surrounded by sound) and imaging that works for both front seats.

3.11 #10 · Find your stage's true edges with the balance control.

With delays and L/R EQ done, pan fully left: that is as far left as the system will image consistently; then fully right. Nothing in a simple stereo system, in a space about half the size of a bathroom, reliably puts images far outside the speakers. [RARE] Sources: DIYMA 2017-08-18

In plain English: Many people chase a stage that stretches beyond the car's doors. He gives a simple test of the real limits. Once the delays and the left/right EQ match are done, turn the balance control fully left: wherever the sound appears is as far left as your system will place images consistently. Then do the same fully right. He says nothing in a simple stereo system, in a space about half the size of a bathroom, reliably puts images far outside the speakers.

3.12 MS-8 auto-EQ, processors and DSP features

What this part is about: The JBL MS-8 was a car-audio processor from his years at Harman that measured the car and tuned itself automatically (an "auto-EQ"). Most of this part relates to that specific product: how it measured and corrected, its quirks and setup facts. The last points cover other auto-tuning products and smaller DSP features.

3.12 #1 · MS-8 measures with a binaural headset at several head positions, at conversation level.

Mics sit at both ears; up to four seats; four measurements per seat (forward twice, toward each mirror), holding still during sweeps. Per position: 3 sweeps × 2 mics = 6 response measurements; for delay, one sweep per channel × 2 mics. Wearable mics let you work the remote while measuring; calibrating too loud degrades results. Product documentation plus his posts. Sources: MS-8 manual (mirror) · DIYMA 2017-06-28

In plain English: This point relates to one product, the JBL MS-8. "Binaural" means two ears: the MS-8's measuring microphones sit on a headset, one at each ear, so it measures roughly what your ears hear. It can calibrate for up to four seats. For each seat it takes four measurements with your head in different positions (facing forward twice, then turned toward each side mirror), and you hold still while it plays its sweeps (tones gliding from low to high pitch). At each position, 3 sweeps × 2 mics gives 6 response measurements; for setting delay it plays one sweep per channel, picked up by both mics. Because the mics are worn, you can work the remote while measuring. He says calibrating too loud makes the results worse; it measures at conversation level. This point draws on the product's documentation plus his own posts.

3.12 #2 · How MS-8's auto-EQ really worked: a correction built from a spatial average.

It took six impulse measurements, converted each to a response (FFT), averaged them, built a minimum-phase version (an average has no phase), converted back to an impulse, and applied 8 filters per channel to successively longer slices of it, so a filter could take almost any shape; the result was convolved with the audio. Corrections apply per speaker location, not per crossover channel. Ulrich Horbach designed it, not Andy. [RARE] [Conflict: his 2011 FAQ described 8 biquad filters per channel that "can do more work than 512 taps"; his 2017 and 2022 descriptions are FIR-like (filters applied to the impulse, then convolution).] Sources: DIYMA 2017-06-28 · DIYMA 2022-02-03 · DIYMA MS-8 FAQ 2011

In plain English: A "spatial average" combines measurements from several positions around the listener's head, so the correction suits the space your head occupies rather than one exact point. Step by step, as described here: the MS-8 took six impulse measurements (see 3.9 #5); turned each into a frequency response using an FFT (a standard calculation that converts a recording over time into levels at each frequency); averaged them; built a "minimum-phase" version, which assigns the most compact possible timing to that averaged response, needed because an average has no phase of its own; turned that back into an impulse; and applied 8 filters per channel to successively longer slices of it, so each correction could take almost any shape. The result was convolved with the audio, meaning applied to the music as one combined filter. Corrections apply per speaker location (for example, the left front), not per crossover channel (for example, just its tweeter). He notes that Ulrich Horbach designed it, not he. The tag flags a conflict: his 2011 FAQ described 8 "biquad" filters per channel (the basic building block of most digital EQs) that "can do more work than 512 taps" (a measure of FIR filter size), while his 2017 and 2022 descriptions sound more like FIR filtering: filters applied to the impulse, then convolution.

3.12 #3 · With an auto-EQ, crossover settings are only suggestions; limit each channel's EQ to its passband.

MS-8's EQ combines with your filters to set the final acoustic slopes, so choose crossovers for safety and output and recalibrate after any change. In other auto-EQs, using the full-range target on every channel makes the EQ fight your crossover slopes. [RARE] Sources: DIYMA MS-8 FAQ 2011 · DIYMA 2021-04-17

In plain English: Normally you'd think the crossover setting alone decides where a speaker stops playing. With the MS-8, he says, the auto-EQ adds its own corrections on top of your crossover filters, and the two together set the final slopes you actually hear. So the crossover settings are only suggestions: choose them for the speaker's safety and for the output you need, and recalibrate after any change. With other auto-EQ systems, limit each channel's EQ to its passband (the range it is meant to play). If you give every channel the full-range target curve, the EQ fights your crossover slopes, trying to boost back what the crossover cut.

3.12 #4 · MS-8 sets sub level from 50–80 Hz before EQ, and that can over-boost the lows.

A 15" in 0.5 cu ft crossed at 40 Hz plays much louder above 40 Hz and fools it; a small sealed box peaking at 50–80 Hz gets turned down. Fixes he gives: a 50 Hz 12 dB low-pass on the sub amp, the 31-band EQ, or (2018) reverse the sub's polarity and set the sub amp's crossover to its highest setting. He'd have changed this algorithm. [RARE] Sources: DIYMA MS-8 FAQ · DIYMA 2018-04-08

In plain English: This is an MS-8 quirk. Before it EQs, the MS-8 sets the subwoofer's overall level from how loud the sub measures between 50 and 80 Hz. That can fool it. His examples: a 15-inch sub in a tiny 0.5 cubic foot (about 14 liter) box, crossed at 40 Hz, still plays much louder above 40 Hz than below, so the MS-8 sets the bass level from that louder 50–80 Hz output and then tries to EQ the deep bass back in; and a small sealed box whose output peaks at 50–80 Hz gets turned down. Either way, the result can be over-boosted lows. The fixes he gives: a 50 Hz low-pass filter with a 12 dB per octave slope on the sub amplifier; the MS-8's 31-band EQ; or, in 2018, reversing the sub's polarity and setting the sub amp's crossover to its highest setting. He says he would have changed this algorithm.

3.12 #5 · MS-8 delay and seat modes.

It averages one impulse from both ear mics, takes the first highest peak, backs up 12 dB and delays every channel to the latest; fronts are not aligned to subs. "Passenger" reverses the delays; "Front/center" removes L/R delays and keeps only crossover-matching ones; "Driver" beats "Front" only in placing midbass below the center's high-pass. Sources: DIYMA 2013 · DIYMA 2018-04-07 · DIYMA 2018-11-16

In plain English: For timing, the MS-8 averages one impulse measurement from the two ear mics, takes the first highest peak, backs up to the point 12 dB below it to estimate the arrival, and delays every channel to match the latest-arriving one. It does not align the front speakers to the subs (compare 3.9 #12). Its seat modes: "Passenger" reverses the delays, mirroring the driver's setting. "Front/center" removes the left/right delays and keeps only the delays that line up speakers sharing a crossover. "Driver" beats "Front" in only one way: placing the midbass below the center speaker's high-pass (the frequency below which the center stops playing). He adds that "Front" leaves the midbass leaning toward the nearer side.

3.12 #6 · Check an auto-EQ with the same kind of measurement it used.

MS-8's head-worn mics give a darker target than a single mic, so a single-mic check won't match; use a multi-position average. Don't add outside correction to its sides and rears (they can't be encoded back into stereo); if you need per-driver delay, put a MiniDSP after MS-8, set its delays, then calibrate. [RARE] Sources: DIYMA 2017-08-05 · DIYMA 2018-04-08 · DIYMA 2011

In plain English: If you check an auto-EQ's result with your own microphone, measure the way it did. The MS-8's mics are worn on your head, so its target comes out "darker" (less treble) than a single-mic measurement would show. A single-mic check therefore won't match what it did; an average of measurements from several positions can confirm it. Don't add your own outside correction to the MS-8's side and rear channels; he says those can't be encoded back into stereo for such correction. If you need separate delays for each driver, put a MiniDSP (a small, inexpensive DSP) after the MS-8, set its delays first, then run the MS-8 calibration.

3.12 #7 · MS-8 setup and service facts.

With a factory radio, learn whether it is fed from the radio or the factory amp; with an aftermarket radio use low-level inputs (fronts to inputs 1–2) and skip input calibration. It may not boot in cold weather because it waits for its slow LCD (its display has its own processor, so it can't run without it). No real custom firmware exists. Random balance shifts usually mean an intermittent input. Sources: DIYMA 2021-12-28 · DIYMA 2022-01-05 · DIYMA 2019-01-13

In plain English: A few practical MS-8 facts. With a factory radio, find out whether the MS-8 is being fed from the radio itself or from the factory amplifier. With an aftermarket radio, use the low-level (line-level) inputs, with the front channels going to inputs 1 and 2, and skip the input calibration step. In cold weather it may not start up, because it waits for its slow LCD screen; the display has its own processor, so the unit can't run without it. No real custom firmware (unofficial replacement software) exists for it. If the balance shifts randomly, that usually means an input with an intermittent (on-and-off) connection.

3.12 #8 · MS-8's purpose and what he'd change.

He hoped it would end the "mythology" that tuning is an art, and stopped wanting a manual back door. In 2016 he'd have replaced L/R delay with phase EQ for two seats, changed the sub-level algorithm, and deepened the 2 kHz dip in the center-channel target. To put the image "on the hood": cut some 500 Hz–2 kHz and adjust center level. Sources: DIYMA 2010 · DIYMA 2016 · DIYMA MS-8 FAQ

In plain English: He hoped the MS-8 would end the "mythology" that tuning is an art, by doing it through measurement. Early on he asked its engineers for a manual "back door" for hand adjustment, but later stopped wanting one. Looking back in 2016, he would have changed three things: replaced left/right delay with phase EQ (all-pass filters, see 3.8 #14–15) so both front seats work; changed the sub-level algorithm (see #4 above); and made the 2 kHz dip in the center channel's target curve deeper. "On the hood" means an image that seems to sit out beyond the windshield. To get it there: cut some energy between 500 Hz and 2 kHz and adjust the center channel's level. He treats an image "on the hood" as a good result to aim for.

3.12 #9 · Auto-EQs need phase, so they build a minimum-phase curve and must cap Q and boost.

That hides regions that aren't minimum phase, so the algorithm protects itself against boosting into nulls. Hand workaround for tools that need phase: paste the phase of a single-mic measurement that resembles the average into the average's file. Averaging phase itself is meaningless (−180° and 0° "average" to −90° but still null). [RARE] Sources: DIYMA 2017-06-27 · DIYMA 2017-06-28 · DIYMA 2017-06-27

In plain English: To design accurate filters, most auto-EQs need phase (timing) information as well as levels. A spatial average has no phase, so they build a minimum-phase version of it (see #2). That hides the parts of the response that aren't minimum phase. Background: deep, narrow dips caused by cancellation (nulls) can't be filled by boosting, because the added sound cancels too. So the software protects itself by capping Q (how narrow a filter can be) and how much it can boost, so it doesn't boost into nulls. His hand workaround for tools that need phase: take a single-mic measurement whose curve looks like the average, and paste its phase data into the average's file. Averaging phase itself is meaningless: two speakers at −180° and 0° "average" to −90°, yet they are still 180° apart and cancel.

3.12 #10 · MS-2, Dirac and APL1 in brief.

After MS-8, Harman found a smoothed single-mic average about 90% as accurate as a spatial average, so MS-2 used one mic and two 512-tap FIRs split at 1 kHz (~1.9 Hz resolution below it). Dirac uses IIR for response plus an FIR working on phase, which improves imaging more than delays can (he hasn't used it and won't rank it). APL1 averages the whole cabin and applies a 4096-tap FIR; repeatability is one test of any automated process. Sources: DIYMA 2017-06-26 · DIYMA 2022-02-03 · DIYMA 2017

In plain English: This point covers three other auto-tuning products. Background: FIR (finite impulse response) and IIR (infinite impulse response) are two kinds of digital filter; FIR filters can shape timing (phase) as well as level, and more "taps" means finer detail. MS-2 was a later Harman processor. After the MS-8, Harman found that a single-mic measurement, smoothed (averaged across neighboring frequencies), was about 90% as accurate as a spatial average, so the MS-2 used one mic. It used two 512-tap FIR filters split at 1 kHz, giving about 1.9 Hz resolution below 1 kHz. Dirac (a commercial correction system) uses IIR filters for the response plus an FIR working on phase, which he says improves imaging more than delays can; he hasn't used it and won't rank it. APL1 averages measurements across the whole cabin and applies a 4096-tap FIR. One test of any automated process, he notes, is repeatability: does it give the same result each time?

3.12 #11 · Check an upmixer with pink noise.

In Logic 7/Pro Logic II, mono noise should come mostly from the center (fronts about 6 dB down); mono noise with one channel inverted comes from the rears; balance hard right plays only the right front; stereo noise exercises everything and is the best overall check. Most DSP makers omit upmixers not for license cost but because product managers don't understand them or judged them from bad setups. [RARE] Sources: DIYMA 2017-08-05 · DIYMA 2017-07-07

In plain English: An upmixer, such as Logic 7 or Dolby Pro Logic II, turns two-channel stereo into more channels: center, sides and rears. His pink-noise checks: mono noise (identical in both channels) should come mostly from the center speaker, with the left and right fronts about 6 dB quieter. Mono noise with one channel's polarity inverted (so the two channels are exact opposites) comes from the rear speakers. With the balance hard right, only the right front plays. Stereo noise (different in each channel) exercises every speaker and is, in his view, the best overall check. He adds that most DSP makers leave upmixers out not because of license costs, but because their product managers don't understand them or judged them from badly set-up systems.

3.12 #12 · Tuning with a binaural head is harder, not easier.

A dummy head's response changes with every direction, so you need either an average that is exact for no direction or a separate calibration file per speaker location (dash, door, center). [RARE] Sources: Dean & Fernando live @58:03

In plain English: A binaural head (dummy head) is a model head with microphones in its ears, meant to measure what a person hears. You might expect it to make tuning easier. He says it makes it harder. A head and ears change the measured response depending on which direction the sound comes from. So you need either an average correction, which is exactly right for no direction, or a separate calibration file for each speaker location (dash, door, center).

3.12 #13 · Smaller DSP notes.

Test claimed "transient" differences between DSPs with a null test (align recordings, invert one, mix; what's left is the difference). Filter arithmetic precision (single vs double, fixed vs floating point) matters. Head-unit control of a Helix fails if the head unit's Wi-Fi is joined to an in-car hotspot. Sources: DIYMA 2021-01-04 · DIYMA 2021-12-18 · DIYMA 2021-09-02

In plain English: Three short notes. First, some people claim one DSP handles "transients" (sudden, sharp sounds such as drum hits) better than another. He says test such claims with a null test: record the same music through both, line the recordings up exactly, invert one, and mix them; whatever is left over is the real difference. Second, the precision of the arithmetic a DSP uses for its filters matters: "single vs double" precision and "fixed vs floating point" are different ways of handling numbers, with different accuracy. Third, Helix is a DSP brand that some head units (car radios) can control; that control fails if the head unit's Wi-Fi is joined to an in-car hotspot.

3.13 Tuning mistakes he sees most

What this part is about: This part lists the tuning errors he says he runs into most often. Many of them point back to rules explained in earlier parts (phase, delay, gain setting, crossovers), and each one shows what goes wrong when a rule is skipped or mixed up.

3.13 #1 · Fixing the center by flipping one mid, then adding delays on top.

Result: midbass 180° out at all frequencies and no center (3.8 #2). Sources: Tuning PDF v2 (~2023) · OFCAP Ep. 19 @3:17:02

In plain English: The "center" is the illusion of a singer standing in the middle of the dashboard, even though the sound comes from left and right speakers. People use two separate fixes for a weak center. One is flipping the polarity of one front speaker (making its cone move backward when the other moves forward). The other is delay: holding back the sound from the nearer speakers so everything arrives at your ears together. He says the mistake is doing both. His point is that you can do one or the other, not both. Once the delays line up the arrival times, the flipped speaker is exactly opposite to the other one, 180° out at every frequency. In the midbass the two sides then cancel, and the center is gone.

3.13 #2 · Entering delays wrongly, then blaming the method.

A competition car with great bass had no center because the owner typed distances into a DSP that wanted delays; "the tape measure doesn't work" is usually this. His rule: if you follow the manual and don't get the result, assume you made an error and keep looking. Forum "14-year breakthroughs" after endless rebuilds usually mean the delays were finally right. [RARE] Sources: Absolute Electronix livestream @44:23 · Absolute Electronix livestream @47:54 · OFCAP Ep. 19 @3:17:02

In plain English: Time alignment means delaying the nearer speakers so that every speaker's sound reaches your ears at the same moment. The usual way is to measure each speaker's distance to your head with a tape measure. Different DSPs (digital sound processors, the box that does the tuning) want that number entered in different ways: as a time, as a distance, or as each speaker's own distance, which you then have to subtract. He describes a competition car with great bass but no center, because the owner typed distances into a DSP that wanted delays. He says that when people claim "the tape measure doesn't work", it is usually this kind of entry mistake. His rule: if you follow the manual and don't get the result, assume you made an error and keep looking. Many people believe their big breakthrough came from years of rebuilding panels and moving speakers. He says those forum "14-year breakthroughs" usually mean the delays were finally set right.

3.13 #3 · Using delay as a pan pot, a depth knob or a crossover-phase fix, or aligning the fronts to the sub.

Delay corrects distance and nothing else (3.9 #1, #8, #9, #12). Sources: Audiofrog blog Dec 2020, Time Alignment Part 1 · DIYMA 2015 · DIYMA 2013

In plain English: A pan pot is a control that slides a sound left or right. A "depth knob" here means trying to push the stage farther forward or back. A crossover splits the music by frequency between speakers, and its filters shift the timing (phase) of the sound near the split point. He says delay has one job: making up for the different distances from each speaker to your ears. Using it to shift the image sideways, to add depth, to cure phase at a crossover, or to line the front speakers up with the subwoofer are all mistakes on his list. The points it refers to explain why (3.9 #1, #8, #9, #12). Note that his advice on delaying the sub itself changed over the years (see 3.9 #12 and 4.2 #12).

3.13 #4 · Listening and tweaking by ear before the objective setup is done, or mixing several teachers' steps.

By-ear tuning means aimlessly fiddling "for 20 minutes"; an amalgam of processes isn't a process ("ignorance is not a process"). Pick one that ends at correct, is teachable and fast, and practise it. Sources: DIYMA 2021-05-12 · OFCAP Ep. 19 @3:19:02 · Audiofrog blog 2021-04-14, UMi-1 CD

In plain English: Tuning has a measured part, done with a microphone (levels, delays, EQ toward a target), and a listening part. He says listening and tweaking by ear before the measured setup is finished turns into aimless fiddling "for 20 minutes". He also says that taking a few steps from one teacher and a few from another doesn't give you a process ("ignorance is not a process"). His advice: pick one process that ends at a correct result, can be taught, and is fast, and then practise it. Elsewhere he says to listen at the end to check the result, not to tweak by ear at every step.

3.13 #5 · Chasing what the mic sees but you don't hear.

EQing the summed response, boosting nulls, and fixing narrow, position-dependent dips (above ~1 kHz, Q above ~6). Move the mic a little: if a ripple stays, fixing it may help; if not, fixing it makes things worse. [RARE] Sources: Tuning PDF v1 (~2018) · Audiofrog blog 2021-01-14, single mic or multi-mic · DIYMA 2021-03-26

In plain English: A measurement mic shows many bumps and dips that your ears don't hear the same way. An EQ (equalizer) raises or lowers chosen frequencies. A null is a deep dip where two sound waves cancel. Q describes how narrow an EQ adjustment is: the higher the Q, the narrower. He lists these as mistakes: EQing the summed response (both sides measured playing together), boosting nulls, and fixing narrow dips that change when the mic moves (above about 1 kHz, with a Q above about 6). Many people believe every wiggle on the graph should be flattened. His test is simple: move the mic a little. If a ripple stays put, fixing it may help. If it moves or disappears, it isn't a real feature of the sound at your head, and fixing it makes things worse.

3.13 #6 · Clipping the DSP's input, then blaming the speakers.

Setting the input range below the source voltage (a 1 V setting on a 4–5 V source allows ~12 dB of A/D clipping) causes permanent distortion that "sounds like broken speakers." He estimates ~70% of GB25 (2.5" midrange) warranty claims are DSP clipping from gain-setting errors; raising the tweeter crossover to "fix distortion" usually hides it (or a leaky pod). Setting gains with sines and a scope before tuning is the related error. [RARE] Sources: Audiofrog blog 2021-03-24, scopes and sine waves · DIYMA 2022-03-13 · Audiofrog blog Dec 2020, pink noise

In plain English: A DSP turns the incoming signal into numbers with an A/D (analog-to-digital) converter. Its input range setting (for example 1 V or 4 V) must match how strong the source's signal is. Set it to 1 V when the radio puts out 4–5 V, and the tops of the waves get chopped off; this is "clipping", by about 12 dB in his example. Because it happens before the processing, the distortion is permanent and nothing later can remove it. It "sounds like broken speakers." He estimates that about 70% of warranty claims on his GB25 (a 2.5-inch midrange) are really DSP clipping from gain-setting errors. People often raise the tweeter's crossover point to "fix" the distortion. He says that usually just hides it, or hides a leaky pod (the small speaker housing not sealed properly). He also calls setting gains with sine-wave tones and an oscilloscope before tuning a related error. So in practice: set the input range to suit your source first, so the DSP's input can't clip, even at full volume.

3.13 #7 · Crossovers by wishful thinking.

Asking a door midbass for 40–60 Hz, wide sub/midbass overlap, mids below Fs, tweeters below ~1.5× Fs, or a steep slope used to justify a low tweeter point. People also mistake a big peak for the speaker's real output: set the target near the dips and cut the peaks instead. Sources: DIYMA 2010 · Audiofrog blog 2021-10-07, slopes and protection · DIYMA 2021-07-12

In plain English: Fs is a speaker's resonant frequency, the pitch where it naturally vibrates most easily; below it, its output falls away. Here he lists crossover choices made on hope rather than on what the speakers can do: asking a door midbass to play deep bass at 40–60 Hz; letting the sub and midbass overlap over a wide range; playing midranges below their Fs; crossing tweeters lower than about 1.5 times their Fs; and counting on a steep slope (a filter that cuts off sharply) to justify crossing a tweeter low. He also says people mistake a big peak on the graph for the speaker's real output level. Instead, set the target level near the dips and cut the peaks down to it.

3.13 #8 · Aiming midranges to fix reflection dips, or boosting a dash speaker's natural treble roll-off.

Sources: Audiofrog blog 2021-06-14, dispersion and reflection

In plain English: Two more mistakes. First, aiming midranges to try to fix the dips caused by reflections (sound bouncing off glass or panels and arriving a moment after the direct sound). In the source article he explains that a midrange sends sound in all directions below a certain frequency, so aiming it does nothing to its reflections; only a tweeter's very top end can be steered by aiming. Second, boosting the treble that a dash speaker naturally loses. He says a dash speaker's steep treble roll-off (its fall in output toward the high notes) shouldn't be treated as a dip to fill: treat what lies below it as a peak and cut that instead.

3.13 #9 · Arguing over whose target curve is best instead of hitting one; or "correcting" a car with a 31-band RTA and 31-band EQ.

Sources: Audiofrog blog 2021-03-10, target curves · DIYMA 2008

In plain English: A target curve is the shape you tune the frequency response toward, for example more bass and gently falling treble. He says people waste time arguing about whose target is best, when the real work is actually hitting one. An RTA (real-time analyzer) shows sound level in bands; a 31-band graphic EQ has 31 fixed sliders. He says this combination's resolution is too coarse to correct a car, though a 31-band EQ is passable for shaping a curve that has already been corrected properly.

3.13 #10 · Ignoring the factory processing already in the signal, or insisting on "flat" first.

Factory all-pass filters can't be "un-EQ'd," so getting "back to zero" first is often inefficient (3.8 #20), and leftover factory delay wrecks yours (3.9 #13). In upmixing factory systems he often keeps the processing: map each factory output to a matching amp channel, add power and better speakers, re-EQ and set levels, with no new delays or crossovers; ripping out factory locations is "a big mistake." [RARE] Sources: Absolute Electronix livestream @58:34 · CMA Connected 2021 @37:25 · CMA Connected 2021 @38:54

In plain English: Factory systems often process the signal before it reaches the speakers: EQ, delays and all-pass filters. An all-pass filter shifts the timing (phase) of a band of frequencies without changing its level, so an EQ cannot undo it. That is why he says trying to get "back to zero" (a flat, unprocessed signal) before tuning is often an inefficient way to work. He also says leftover factory delay will spoil any delays you add on top. Many people believe the first step is always to flatten the factory signal. In factory systems that "upmix" (spread stereo music over many speakers), he often keeps the factory processing. He connects each factory output to a matching amp channel, adds power and better speakers, re-EQs and sets levels, with no new delays or crossovers. He calls ripping out the factory speaker locations "a big mistake."

3.13 #11 · Four months of fabrication, 20 minutes of tuning.

Shops show him beautiful builds that don't sound great and say "not bad for the 20 minutes I spent"; the audio deserves real time. [RARE] Sources: CMA Connected 2021 @22:22

In plain English: Fabrication means custom-building panels, pods and boxes. He says shops show him beautiful builds that don't sound great, with remarks like "not bad for the 20 minutes I spent" on tuning. His point: if the build took months, the sound deserves real tuning time too.

3.13 #12 · Treating customers who don't care about imaging as not needing a DSP.

Club-style listeners need it for driver safety and for midbass-to-sub phase, which is where impact comes from (3.8 #10). Sources: CMA Connected 2021 @46:57

In plain English: Imaging is hearing each instrument at its own spot on a stage in front of you. Some customers only want loud, punchy, club-style sound and don't care about that. He says they still need a DSP, for two reasons: keeping the speakers (drivers) safe, and getting the midbass and subwoofer in phase, which he says is where impact comes from (3.8 #10). Note that 4.2 #3 records a different statement of his about where "impact" comes from.

3.13 #13 · Tearing an install apart for hiss that the auto-EQ created.

If a factory head unit rolls off the highs, an auto-EQ boosts the top end to compensate, and that boost is the hiss; it isn't a ground fault. [RARE] Sources: DIYMA 2021-08-08

In plain English: Hiss is a steady background "shhh". Auto-EQ is a feature that measures the car and sets the EQ automatically. If the factory head unit (the radio) turns down the high frequencies, the auto-EQ boosts the top end to make up for it, and that boost is what you hear as hiss. He says it isn't a ground fault (a bad electrical connection to the car's body), so there is no point tearing the install apart looking for one. Many people assume hiss always means a grounding problem.

Top rarely-discussed points in 3.8–3.13

1. An all-pass on one side fixes the ~250 Hz center problem for both front seats; delay fixes one and doubles the other's error (3.8 #14, 3.9 #14)

2. Two quick tests reveal factory all-pass vs delay: a single 200–500 Hz dip in L+R vs a comb (3.8 #17)

3. Keep factory polarity and add no delays on top of factory all-pass filters; they can't be inverted (3.8 #20)

4. Read phase from the L+R sum: +6 dB in phase, +3 dB ≈ 90°, equal to one side ≈ 120° (3.8 #7)

5. In-phase crossover pairs are heard at the higher-frequency driver; that's how bass comes forward (3.8 #10)

6. Delay doesn't move speakers; center with level, never delay (3.9 #1, 3.10 #2)

7. Impulse peaks aren't arrival times for subs and midbass; a tape measure is as good or better (3.9 #5)

8. Auto time alignment measures dash speakers via the windshield (3.9 #6)

9. Reflections and cabin gain never change delay settings (3.9 #10)

10. Learn imaging from a matched $5 thrift-store pair; "in phase" is the first reference (3.11 #2)

11. Seven-drum-beat and vocal tracks can't check midbass; use panned bass guitar (3.11 #4)

12. ~70% of GB25 (2.5" midrange) warranty claims are DSP input clipping from gain-setting errors, not bad drivers (3.13 #6)

Changed views and conflicts in 3.8–3.13

• All-pass phase EQ (3.8 #15): 2010, never got one to work for sub/midbass blending; 2014–15, Harman tests found "no audible difference"; from 2016, his preferred two-seat tool, and by 2021–22 the reason not to add delays to factory systems.

• Aligning fronts to the sub (3.9 #12): he argued for it at Harman, then accepted trials showing no difference below 100 Hz; the Tuning PDFs still tape-measure the sub like any speaker.

• The passenger seat (3.9 #14): once a trade-off against the driver's seat while delay was the only tool; now he calls "forget the passenger" a cop-out, since all-pass filters, placement or a center fix both seats.

• Rear fill (3.9 #18): 2017 recipes for differential L−R fill (delayed 12–20 ms); from late 2021, plain in-phase stereo delayed 10–20 ms.

• MS-8 filter type (3.12 #2): 2011 "8 biquads"; 2017 and 2022 an FIR-like correction convolved with the audio.

• Test-track numbering (3.11 #3–4): the 2021 UMI-1 video and the Tuning PDFs number the band-limited noise tracks differently.

• "7 drum beats" (3.11 #4): he says it can't check midbass; the attributed gain guide suggests it for centering.

---

The full series

1. Start here

2. Part 1: How Stereo Imaging Works in a Car, Reflections and Dispersion

3. Part 2: Aiming and Where Each Speaker Goes

4. Part 3: One Seat vs Two, Center Channels, Rear Fill and Upmixing

5. Part 4: Door and Midbass Installation, Sound Deadening and Speaker Locations

6. Part 5: Pods, Tweeter Mounting and Choosing Drivers

7. Part 6: Speaker Design, Driver Failures, Tweeter Protection and Wiring

8. Part 7: His Step-by-Step Tuning Process and Measurement Tools

9. Part 8: Mic Placement, Target Curves and EQ

10. Part 9: Crossovers, Polarity and Phase

11. Part 10: Time Alignment, Levels, Listening Tests, Processors and Common Tuning Mistakes

12. Part 11: Bass in a Car: Cabin Gain, Front Bass, Enclosures and Subwoofer Placement

13. Part 12: Power, Clipping, Gain Setting, Amplifier Specs and Noise

14. Part 13: Wiring and Integrating with Factory Systems

15. Part 14: Car Audio Myths He Debunks (1 of 2)

16. Part 15: Car Audio Myths He Debunks (2 of 2)

17. Part 16: Reading Specs and Reviews, and Planning a System

18. Part 17: Choosing a Shop, the Car Audio Business and SQ Competitions

19. Part 18: Recordings, Listening and Who Andy Wehmeyer Is

20. Part 19: Glossary and Where His Material Lives

Download the complete guide as one PDF (351 pages, every point with its explanation).