Realtime adjuster results
LevelAPI2 202610170, Medium preset, in each of its three modes (50 ms, 2 ms, and 2 ms holding the dialogue), measured by the published test method on 117 cases at -23 LUFS and 68 programme cases at -16 LUFS: synthetic level jumps and pauses, bass tones, broadcast recordings, films, music, read speech, conversation and a live stream: 25 hours of material. Every graded line is shown with every case that fails it and the measured figure.
| Declaration | Value |
|---|---|
| Device | LevelAPI2 realtime adjuster, revision 202610170, preset Medium, every other setting at its default, in each mode of RealtimeAdjuster_SetMode(): 0, standard; 1, low latency; 2, low latency with dialogue |
| Target | -23 LUFS; the programme cases again at -16 LUFS |
| True-peak ceiling | -1 dBTP, the default (SetTruePeakCeilingDb) |
| Latency | 50 ms in mode 0, 2 ms in modes 1 and 2, measured on the audio: mode 0: 2400 frames at 48 kHz, 2205 at 44.1 kHz; mode 1: 96 frames at 48 kHz, 88 at 44.1 kHz; mode 2: 96 frames at 48 kHz, 88 at 44.1 kHz |
| Dialogue anchor | Mode 2: the calibration holds the dialogue's loudness rather than the programme's, both read by the library's speech detector. Below 10 % speech it holds the programme, above 25 % the dialogue, and blends between |
| Maximum boost | +12 dB |
| Reference block | 10 ms (480 frames at 48 kHz, 441 at 44.1 kHz) |
| Block sizes declared identical | every size, in every mode. Tested at 20 ms and 256, 512, 1024 and 2048 frames on every case but P6 and the conversation cases (108 cases in mode 0, 112 in modes 1 and 2, with the tones), and at 64 frames on the synthetic cases and the EBU / ITU files (49 in mode 0, 53 in modes 1 and 2) |
| Program length | the whole stream by default; the long cases again with 900 s |
The method's pass lines, a column per mode. For mode 2, which holds the dialogue at the target, G1-D takes the place of G1 and G2, which are reported beside it; modes 0 and 1 report G1-D. A failing row stays on this page with its figure and what it is.
| Pass line | Scope | Mode 0, standard | Mode 1, low latency | Mode 2, low latency with dialogue |
|---|---|---|---|---|
| G1 | Whole-case integrated output within ±1 LU of ‑23 LUFS, every programme case | 68 of 68; 15 of 15 with a 900 s program length; the same at every block size | 68 of 68; 15 of 15 with a 900 s program length; the same at every block size | reported: 61 of 68; 15 of 15 with a 900 s program length; the same at every block size |
| G1-D | Output dialogue loudness within ±1 LU of the target, on the programme cases whose input is at least 25 % speech, both read by the library's speech detector | reported: 43 of 45 | reported: 44 of 45 | 45 of 45 |
| G2 | Long-term loudness in band on every case longer than 16 min | 13 of 15; 14 of 15 with a 900 s program length | 12 of 15; 13 of 15 with a 900 s program length | reported: 10 of 15; 9 of 15 with a 900 s program length |
| G3 | Maximum true peak at or under ‑0.8 dBTP on the lossless cases (the synthetic signals, the tones and the EBU / ITU files), at every block size | 51 of 53; the highest passing case reads ‑0.99 dBTP | 53 of 53; the highest passing case reads ‑1.45 dBTP | 53 of 53; the highest passing case reads ‑1.59 dBTP |
| G4 | The last 30 s of each constant synthetic part within ±1 LU, where the need is within the +12 dB maximum | 67 of 76, the same parts at every block size | 67 of 76 | 68 of 76 |
| G5 | The output at every other block size identical to 10 ms blocks, sample for sample | 588 of 589 runs at 6 other sizes | 612 of 613 runs at 6 other sizes | 612 of 613 runs at 6 other sizes |
| Line | Case | Mode | Material | Measured | What it is |
|---|---|---|---|---|---|
| G2 | P12 | 0, 1 | Live stream 1, game audio and voice | 0: 87.03 % in band 1: 88.10 % in band | A sparse live stream: over its quiet stretches (-35 to -58 LUFS) the gain sits on the +12 dB maximum and the output stays low. The fixed-gain two-pass reference reads 69.1 % here. |
| G2 | P13.3 | 1 | Conversation 3, panel discussion, TIB AV-Portal (CC BY 3.0 DE) | 91.63 % in band | An 84-minute panel discussion recorded about 6 LU under the target. For about ten minutes the gain sits on the +12 dB maximum and the long-term loudness on the lower edge of the band. The 2 ms peak stages take about 0.15 dB more off this speech than the 50 ms mode, and on the maximum the calibration cannot give it back. |
| G2 | P6 | 0, 1 | Long run, four mixes back to back (workout, mixed, pop, rock) | 0: 92.24 % in band 1: 89.36 % in band | Three hours and 42 minutes of music with no program length: after two hours a new minute is too small a share of the long-term correction to move it. With 900 s it passes. |
| G3 | S6.1 | 0 | pink T-15, 10 s silence, bursts at T | 10 ms ‑0.46, 20 ms ‑0.46, 64 frames ‑0.46, 256 frames ‑0.46, 512 frames ‑0.46, 1024 frames ‑0.46, 2048 frames ‑0.46 dBTP | White-noise bursts after a quiet passage and digital silence. The limiter holds -1 dBTP on its own 4x estimate; a high-precision true-peak meter can read white-noise bursts up to about 1.7 dB higher (0.54 dB on this case). The ceiling setting is there for this: set it that much lower for a guarantee on such material. |
| G3 | S6.2 | 0 | pink T-15, 10 s room tone at -50, bursts at T | 10 ms ‑0.46, 20 ms ‑0.46, 64 frames ‑0.46, 256 frames ‑0.46, 512 frames ‑0.46, 1024 frames ‑0.46, 2048 frames ‑0.46 dBTP | As S6.1, with room tone at -50 LUFS instead of silence. |
| G4 | S3.up part 2 | 0, 1 | P5, -12 dB -> 0 dB at 180 s | 0: ‑24.89 LUFS 1: ‑24.08 LUFS | Programme material split at its file joins, 30 to 83 s after a 12 dB step: the ride has not settled yet. |
| G4 | S3.up part 3 | 0, 1, 2 | P5, -12 dB -> 0 dB at 180 s | 0: ‑20.50 LUFS (the fixed-gain reference reads ‑20.50) 1: ‑20.62 LUFS (the fixed-gain reference reads ‑20.50) 2: ‑19.03 LUFS (the fixed-gain reference reads ‑20.50) | Programme material split at its file joins, 30 to 83 s after a 12 dB step: the ride has not settled yet. |
| G4 | S3.up part 4 | 0, 1, 2 | P5, -12 dB -> 0 dB at 180 s | 0: ‑24.80 LUFS 1: ‑24.48 LUFS 2: ‑21.93 LUFS | Programme material split at its file joins, 30 to 83 s after a 12 dB step: the ride has not settled yet. |
| G4 | S3.down part 1 | 0, 1, 2 | P5, 0 dB -> -12 dB at 180 s | 0: ‑21.25 LUFS 1: ‑21.04 LUFS 2: ‑21.62 LUFS | As S3.up, after a 12 dB step down. |
| G4 | S3.down part 2 | 0, 1, 2 | P5, 0 dB -> -12 dB at 180 s | 0: ‑25.20 LUFS 1: ‑25.13 LUFS 2: ‑24.51 LUFS | As S3.up, after a 12 dB step down. |
| G4 | S5.-60.30 part 3 | 0, 1, 2 | pink T-10, 30 s room tone at -60, pink T-10 | 0: ‑24.13 LUFS 1: ‑24.13 LUFS 2: ‑24.13 LUFS | Room tone in a pause of 30 s or more is lifted up to 12 dB. The returning programme meets that boost and is over-corrected afterwards, so the part ends up to 1.6 LU low. |
| G4 | S5.-60.120 part 3 | 0, 1, 2 | pink T-10, 120 s room tone at -60, pink T-10 | 0: ‑24.56 LUFS 1: ‑24.56 LUFS 2: ‑24.56 LUFS | Room tone in a pause of 30 s or more is lifted up to 12 dB. The returning programme meets that boost and is over-corrected afterwards, so the part ends up to 1.6 LU low. |
| G4 | S5.-50.120 part 3 | 0, 1, 2 | pink T-10, 120 s room tone at -50, pink T-10 | 0: ‑24.26 LUFS 1: ‑24.26 LUFS 2: ‑24.26 LUFS | Room tone in a pause of 30 s or more is lifted up to 12 dB. The returning programme meets that boost and is over-corrected afterwards, so the part ends up to 1.6 LU low. |
| G4 | S5.-45.120 part 3 | 0, 1, 2 | pink T-10, 120 s room tone at -45, pink T-10 | 0: ‑24.07 LUFS 1: ‑24.07 LUFS 2: ‑24.07 LUFS | Room tone in a pause of 30 s or more is lifted up to 12 dB. The returning programme meets that boost and is over-corrected afterwards, so the part ends up to 1.6 LU low. |
| G5 | S7 at 2048 frames | 0 | pink T-10, reset at 120 s | 85 of 3000 ticks differ from 120.0 s | The test harness resets the stream on the first host block boundary at or after 120 s. At 2048 frames that boundary falls 1024 frames later than at every other size, so the stream is reset at a different sample. Identical again from 140.8 s. |
| G5 | S7 at 2048 frames | 1, 2 | pink T-10, reset at 120 s | 1: 81 of 3000 ticks differ from 120.0 s 2: 81 of 3000 ticks differ from 120.0 s | The test harness resets the stream on the first host block boundary at or after 120 s. At 2048 frames that boundary falls 1024 frames later than at every other size, so the stream is reset at a different sample. |
Where a mode's figure is outside a line that is reported for it rather than graded.
| Line | Case | Mode | Material | Measured | What it is |
|---|---|---|---|---|---|
| G1 | P4.1 | 2 | ITU BS.1770-2 conformance, stereo | ‑20.75 LUFS (dialogue ‑22.95) | An ITU conformance piece whose dialogue sits under the mix: mode 2 holds the dialogue at the target, so the programme sits above it. |
| G1 | P4.2 | 2 | ITU BS.1770-2 conformance, mono | ‑21.12 LUFS (dialogue ‑23.16) | An ITU conformance piece whose dialogue sits under the mix: mode 2 holds the dialogue at the target, so the programme sits above it. |
| G1 | P9.01 | 2 | Agent 327 (Blender, CC-BY) | ‑21.49 LUFS (dialogue ‑22.34) | A film with little speech, where mode 2 blends the programme and the dialogue; the dialogue sits under the mix, so the programme lands above the target. |
| G1 | P9.08 | 2 | Elephants Dream (Blender, CC-BY) | ‑21.41 LUFS (dialogue ‑22.97) | A film with little speech, where mode 2 blends the programme and the dialogue; the dialogue sits under the mix, so the programme lands above the target. |
| G1 | P9.14 | 2 | Sprite Fright (Blender, CC-BY) | ‑21.94 LUFS (dialogue ‑22.85) | A film whose dialogue sits under its mix: mode 2 holds the dialogue at the target, so the programme sits above it. |
| G1 | P9.16 | 2 | The Daily Dweebs (Blender, CC-BY) | ‑21.67 LUFS (dialogue ‑22.53) | A film whose dialogue sits under its mix: mode 2 holds the dialogue at the target, so the programme sits above it. |
| G1 | P12 | 2 | Live stream 1, game audio and voice | ‑24.17 LUFS (dialogue ‑22.54) | A game stream whose voice sits well above the game audio: mode 2 holds the voice at the target, so the programme sits under it. |
| G1-D | P9.16 | 0 | The Daily Dweebs (Blender, CC-BY) | dialogue ‑24.55 LUFS (programme ‑22.48) | A film whose dialogue sits under its mix: with the programme at the target, the dialogue sits under it. |
| G1-D | P12 | 0, 1 | Live stream 1, game audio and voice | 0: dialogue ‑20.98 LUFS (programme ‑22.87) 1: dialogue ‑21.64 LUFS (programme ‑22.96) | A game stream whose voice sits well above the game audio: with the programme at the target, the voice sits above it. |
| G2 | P10.3 | 2 | Music mix 3, pop | 92.45 % in band | A music mix in which the speech detector hears speech. Mode 2 then holds that speech at the target; it sits under the mix, so the music is lifted, for tens of minutes. Mode 1 is the mode for music. |
| G2 | P10.5 | 2 | Music mix 5, workout | 57.65 % in band | A music mix in which the speech detector hears speech. Mode 2 then holds that speech at the target; it sits under the mix, so the music is lifted, for tens of minutes. Mode 1 is the mode for music. |
| G2 | P12 | 2 | Live stream 1, game audio and voice | 23.86 % in band | A game stream whose voice sits well above the game audio: mode 2 holds the voice at the target, so the programme sits under it. |
| G2 | P13.3 | 2 | Conversation 3, panel discussion, TIB AV-Portal (CC BY 3.0 DE) | 93.77 % in band | An 84-minute panel discussion recorded about 6 LU under the target. For about ten minutes the gain sits on the +12 dB maximum and the long-term loudness on the lower edge of the band. The 2 ms peak stages take about 0.15 dB more off this speech than the 50 ms mode, and on the maximum the calibration cannot give it back. |
| G2 | P6 | 2 | Long run, four mixes back to back (workout, mixed, pop, rock) | 84.31 % in band | The 3 h 42 min feed opens with the music of P10.5, which mode 2 lifts; with no program length the long-term loudness carries that start for hours. |
Where the adjuster is reliable, and what it does outside that range, with the modes each row concerns. "What changes it" names only what the shipped API offers, and what a host controls anyway.
Three excerpts of real programme, to hear in both 2 ms modes against the original, are on the listening examples page.
| What the listener hears | Case that shows it | Modes | What changes it |
|---|---|---|---|
| After a jump to louder material (an ad after a quiet programme) it is too loud for about half a minute, by up to 7.7 LU, then settles | S1 and S2 steps up: back in band in 26-48 s (S2.20, which arrives on the +12 dB maximum: 85 s), in mode 0; mode 1 is never louder over 3 s | all | Not adjustable. A host that knows the boundary (a playlist change, an ad insertion) can call Reset() there: the stream starts fresh from 0 dB (S7: back in band 9.8 s after a reset). Reset() allocates, so call it off the audio thread |
| After a drop to quieter material it stays quiet for up to a minute and a half | S1 and S2 steps down: back in band in 47-83 s | all | Not adjustable; Reset() at a known boundary as above |
| After a long loud passage (an explosion in a film), the quieter programme that follows can sit more than 6 LU under the target for tens of seconds. Modes 1 and 2 remove 63 % and 52 % of it | Every case at ‑23 LUFS, depth-weighted: 762 / 283 / 368 LU-seconds in modes 0 / 1 / 2; the longest, P9.12, 61 s in mode 0. Counted where the input is audible, the output more than 6 LU under the target and 3 LU under the fixed-gain reference, after a passage at least 8 LU louder in the last 30 s | all | Modes 1 and 2 |
| Material more than 12 LU under the target stays under it | S1.-20 (never back in band), P12's quiet stretches | all | Not adjustable: +12 dB is the most the adjuster adds |
| At a raised target, quiet material stays under it: at -16 LUFS, anything quieter than about -28 LUFS | -16 LUFS run: P7.02 (‑17.65 LUFS), P13.3 (‑18.79 LUFS) | all | SetTargetLUFS(): a lower target leaves more material within +12 dB. The same cases pass at -23 LUFS |
| A quiet talk on the +12 dB maximum sits a little lower in the 2 ms modes | P13.3, a panel discussion: long-term loudness in band 91.6 % in mode 1, 96.8 % in mode 0 | 1, 2 | A lower target, as in the row above |
| Background noise in a pause is raised (room tone, hiss, an audience), by up to 12 dB when the pause lasts 30 s or more. The programme then returns about 2 LU loud for 25-51 s, then up to 1.6 LU quiet | S5, room tone for 30 s or 120 s. Five seconds of room tone costs nothing | all | Not adjustable |
| After digital silence: pauses of 10 s or less return unchanged. After 30 s, material that needed a lift returns up to 4 LU quiet for about 6 s; after a minute or more the correction has faded back to 0 dB, and quiet material returns up to 10 LU under the target, back in band in about 12 s | S4, silence of 2 to 120 s, in the 50 ms mode. Material at the target returns clean after every length | all | Not adjustable. A long pause is treated as the end of the programme |
| A stream switched on before its programme starts raises the room tone by 12 dB; the programme arrives up to 7.4 LU loud, then dips, back in band in 24-141 s | S8, room tone for 30 to 300 s before the programme, in the 50 ms mode | all | Start the stream, or call Reset(), when the programme starts |
| A percussive entry after a quiet passage is up to 3.8 LU loud for about 18 s | S6.1, S6.2, in the 50 ms mode | all | Not adjustable |
| Peaks read above the ceiling on a high-precision true-peak meter: up to about 1.7 dB on noise-like material, up to about 0.35 dB on lossy programme at -16 LUFS. In the 2 ms modes the limiter's estimate also covers the point between two samples, and the lossless cases stay under ‑0.8 dBTP | S6: ‑0.46 dBTP in mode 0 at the default -1 dBTP, ‑1.59 in modes 1 and 2. At -16 LUFS: P13.4, ‑0.66 dBTP | all | SetTruePeakCeilingDb(): set the ceiling that much lower. Lossless programme material stays under -0.8 dBTP at the default |
| At a raised target, a bass line made of hard kicks lands under it in the 2 ms modes: their short limiter holds every kick | L2.kick, a 50 Hz kick loop, at -16 LUFS: ‑17.32 LUFS in modes 1 and 2, ‑15.67 in mode 0; the fixed-gain two-pass reference reaches ‑17.42. At -23 LUFS: ‑22.23 / ‑23.00 / ‑23.00 in modes 0 / 1 / 2 | 1, 2 | Mode 0 for very peaky bass at a high target, or a lower target |
| In mode 2, music in which the speech detector hears speech is lifted: its 5-minute loudness sits more than 1 LU above the target for 24 min, up to 2.2 LU | P10.5, a 1 h 00 min music mix (mode 1: 27 s above) | 2 | Mode 1 for music |
In mode 2, the first minute after a start or Reset() settles less evenly: at some point its gain moves up to 6.0 dB further than mode 1's on speech. Over the first minute it sits as close to the target as the 50 ms mode: 3.3 LU from it on average, against 3.3 in mode 0 and 3.0 in mode 1. On the synthetic signals the difference from mode 1 is up to 3.5 dB, for at most 20 s | The 47 speech cases (at least 25 % speech), 10-60 s after the input is first audible | 2 | Not adjustable; mode 1 settles more evenly |
| On a feed that runs for hours, the level drifts up to about 1.3 LU after the first two hours | P6: long-term loudness in band 92.2 %, mode 0 | all | SetProgramLengthSeconds(900): 99.9 % on the same feed; neutral on an hour of material |
| Peaks above the ceiling with the limiter off | not measured | all | SetLimiterEnabled(true), the default |
Median share of the time the output's loudness sits within ±1 LU of the target, over 10 s and 300 s windows, for modes 0 / 1 / 2 / a two-pass normalisation that knows the whole file in advance / the untouched input; and the median gain movement for modes 0 / 1 / 2. The two-pass figure is what a file-based workflow achieves; a live adjuster cannot know the future, and trades that for following the programme as it changes.
| Material | Cases | W10 in band % | W300 in band % | Gain movement dB/min |
|---|---|---|---|---|
| Broadcast recordings | 11 | 77 / 77 / 68 / 66 / 2 | 100 / 100 / 100 / 100 / 0 | 4.0 / 4.0 / 4.2 |
| Zapping stream | 1 | 62 / 62 / 62 / 25 / 29 | 98 / 98 / 97 / 30 / 22 | 3.6 / 3.6 / 3.6 |
| EBU / ITU reference programme | 4 | 24 / 38 / 9 / 11 / 13 | 100 / 100 / 38 / 60 / 59 | 15.1 / 15.4 / 13.8 |
| Films | 17 | 23 / 26 / 24 / 19 / 3 | 94 / 100 / 81 / 63 / 0 | 17.0 / 15.6 / 15.6 |
| Music mixes | 5 | 58 / 58 / 49 / 53 / 0 | 95 / 93 / 89 / 94 / 0 | 2.5 / 2.5 / 2.6 |
| Read speech, long | 3 | 60 / 60 / 61 / 51 / 3 | 100 / 100 / 100 / 91 / 0 | 5.4 / 5.3 / 5.4 |
| Poetry readings, short | 20 | 60 / 62 / 59 / 62 / 0 | 100 / 100 / 100 / 100 / 0 | 5.6 / 7.2 / 4.4 |
| Conversation | 4 | 58 / 59 / 59 / 54 / 1 | 100 / 100 / 100 / 77 / 0 | 5.0 / 5.2 / 5.2 |
| Monologue over a bed | 1 | 55 / 54 / 54 / 37 / 3 | 99 / 98 / 94 / 57 / 1 | 4.4 / 4.6 / 4.7 |
| Live stream | 1 | 23 / 29 / 23 / 22 / 19 | 91 / 88 / 46 / 69 / 31 | 20.6 / 22.2 / 17.9 |
| Long run, 3 h 42 min | 1 | 56 / 55 / 54 / 31 / 1 | 85 / 84 / 75 / 37 / 0 | 2.8 / 2.8 / 2.7 |
Pink noise at defined levels, programme material stepped by 12 dB, silence, room tone and noise bursts. "Need" is the gain a perfect fixed gain would apply after the change; "over" and "under" are the largest excess of the short-term loudness over the minute after it. Each figure is mode 0 / mode 1.
| Case | Signal | Change at s | Need dB | Over LU | Under LU | Above band s | Back in band s |
|---|---|---|---|---|---|---|---|
| S1.+6 | pink T -> T+6 | 180 | ‑6.0 | 6.0 / 6.0 | 0.0 / 0.0 | 27.0 / 26.9 | 35.2 / 35.1 |
| S1.+12 | pink T-2 -> T+10 (re-based) | 180 | ‑10.0 | 7.3 / 5.5 | 0.1 / 0.0 | 42.1 / 44.3 | 42.6 / 45.2 |
| S1.+20 | pink T-10 -> T+10 (re-based) | 180 | ‑10.0 | 7.7 / 6.0 | 0.5 / 0.3 | 44.9 / 54.0 | 47.9 / 56.1 |
| S1.-6 | pink T -> T-6 | 180 | +6.0 | 0.0 / 0.0 | 5.4 / 5.4 | 0.0 / 0.0 | 50.2 / 50.2 |
| S1.-12 | pink T -> T-12 | 180 | +12.0 | 0.0 / 0.0 | 11.8 / 11.8 | 0.0 / 0.0 | 71.1 / 71.2 |
| S1.-20 | pink T -> T-20 | 180 | +20.0 | 0.0 / 0.0 | 20.0 / 20.0 | 0.0 / 0.0 | not in the case / not in the case |
| S2.6 | pink T-6 -> T | 180 | +0.0 | 5.5 / 5.5 | 0.4 / 0.4 | 24.8 / 24.8 | 26.2 / 26.2 |
| S2.12 | pink T-12 -> T | 180 | +0.0 | 7.1 / 5.7 | 0.7 / 0.5 | 38.4 / 40.4 | 39.2 / 42.7 |
| S2.20 | pink T-20 -> T | 180 | +0.0 | 7.2 / 3.8 | 7.1 / 7.3 | 18.2 / 15.0 | 84.9 / 30.2 |
| S2.-6 | pink T+6 -> T | 180 | +0.0 | 0.0 / 0.0 | 5.3 / 5.3 | 0.0 / 0.0 | 47.0 / 47.0 |
| S2.-12 | pink T+10 -> T-2 (re-based) | 180 | +2.0 | 0.4 / 0.4 | 11.4 / 11.4 | 0.0 / 0.0 | 68.3 / 70.1 |
| S2.-20 | pink T+10 -> T-10 (re-based) | 180 | +10.0 | 0.4 / 0.4 | 20.4 / 20.4 | 0.0 / 0.0 | 83.4 / 83.4 |
| S3.up | P5, -12 dB -> 0 dB at 180 s | 180 | ‑1.6 | 4.8 / 3.0 | 2.5 / 2.6 | 32.3 / 34.1 | not in the case / 49.8 |
| S3.up | P5, -12 dB -> 0 dB at 180 s | 246 | +0.0 | 3.9 / 3.9 | 2.2 / 1.7 | 3.7 / 3.8 | not in the case / not in the case |
| S3.up | P5, -12 dB -> 0 dB at 180 s | 330 | +2.9 | 0.0 / 0.0 | 4.8 / 4.9 | 0.0 / 0.0 | not in the case / not in the case |
| S3.down | P5, 0 dB -> -12 dB at 180 s | 180 | +10.4 | 1.7 / 2.4 | 8.3 / 7.8 | 4.5 / 5.5 | not in the case / not in the case |
| S3.down | P5, 0 dB -> -12 dB at 180 s | 246 | +12.0 | 0.0 / 0.0 | 1.1 / 1.2 | 0.4 / 0.4 | not in the case / not in the case |
| S3.down | P5, 0 dB -> -12 dB at 180 s | 330 | +14.8 | 0.0 / 0.0 | 4.8 / 4.4 | 1.0 / 0.0 | not in the case / not in the case |
| S4.T.2 | pink T, 2 s silence, pink T | 122 | +0.0 | 0.0 / 0.0 | 0.0 / 0.0 | 0.0 / 0.0 | 0.1 / 0.1 |
| S4.T.10 | pink T, 10 s silence, pink T | 130 | +0.0 | 0.0 / 0.0 | 0.0 / 0.0 | 0.0 / 0.0 | 0.8 / 0.8 |
| S4.T.30 | pink T, 30 s silence, pink T | 150 | +0.0 | 0.0 / 0.0 | 0.0 / 0.0 | 0.0 / 0.0 | 0.8 / 0.8 |
| S4.T.60 | pink T, 60 s silence, pink T | 180 | +0.0 | 0.0 / 0.0 | 0.0 / 0.0 | 0.0 / 0.0 | 0.8 / 0.8 |
| S4.T.120 | pink T, 120 s silence, pink T | 240 | +0.0 | 0.5 / 0.0 | 0.0 / 0.0 | 0.0 / 0.0 | 0.8 / 0.8 |
| S4.L.2 | pink T-10, 2 s silence, pink T-10 | 122 | +10.0 | 0.0 / 0.0 | 0.5 / 0.5 | 0.0 / 0.0 | 0.1 / 0.1 |
| S4.L.10 | pink T-10, 10 s silence, pink T-10 | 130 | +10.0 | 0.0 / 0.0 | 0.5 / 0.5 | 0.0 / 0.0 | 0.8 / 0.8 |
| S4.L.30 | pink T-10, 30 s silence, pink T-10 | 150 | +10.0 | 0.5 / 0.5 | 4.0 / 4.0 | 0.0 / 0.0 | 6.3 / 6.3 |
| S4.L.60 | pink T-10, 60 s silence, pink T-10 | 180 | +10.0 | 1.0 / 1.0 | 10.0 / 10.0 | 0.0 / 0.0 | 11.9 / 11.9 |
| S4.L.120 | pink T-10, 120 s silence, pink T-10 | 240 | +10.0 | 1.0 / 1.0 | 10.0 / 10.0 | 0.0 / 0.0 | 11.9 / 11.9 |
| S5.-60.5 | pink T-10, 5 s room tone at -60, pink T-10 | 125 | +10.0 | 0.0 / 0.0 | 0.5 / 0.5 | 0.0 / 0.0 | 0.1 / 0.1 |
| S5.-60.30 | pink T-10, 30 s room tone at -60, pink T-10 | 150 | +10.0 | 2.0 / 2.0 | 0.1 / 0.1 | 27.7 / 27.9 | 11.1 / 11.2 |
| S5.-60.120 | pink T-10, 120 s room tone at -60, pink T-10 | 240 | +10.0 | 2.0 / 2.0 | 1.0 / 1.0 | 50.8 / 50.9 | not in the case / not in the case |
| S5.-50.5 | pink T-10, 5 s room tone at -50, pink T-10 | 125 | +10.0 | 0.0 / 0.0 | 0.5 / 0.5 | 0.0 / 0.0 | 0.1 / 0.1 |
| S5.-50.30 | pink T-10, 30 s room tone at -50, pink T-10 | 150 | +10.0 | 2.0 / 2.0 | 0.5 / 0.5 | 29.2 / 29.2 | 38.7 / 38.7 |
| S5.-50.120 | pink T-10, 120 s room tone at -50, pink T-10 | 240 | +10.0 | 2.0 / 2.0 | 1.0 / 1.0 | 40.9 / 40.9 | 42.9 / 42.9 |
| S5.-45.5 | pink T-10, 5 s room tone at -45, pink T-10 | 125 | +10.0 | 0.0 / 0.0 | 0.5 / 0.5 | 0.0 / 0.0 | 5.7 / 5.8 |
| S5.-45.30 | pink T-10, 30 s room tone at -45, pink T-10 | 150 | +10.0 | 2.0 / 2.0 | 0.3 / 0.3 | 25.1 / 24.9 | 31.3 / 31.1 |
| S5.-45.120 | pink T-10, 120 s room tone at -45, pink T-10 | 240 | +10.0 | 2.0 / 2.0 | 0.8 / 0.8 | 35.8 / 35.8 | 42.0 / 42.0 |
| S6.1 | pink T-15, 10 s silence, bursts at T | 130 | ‑0.0 | 3.8 / 2.3 | 1.8 / 0.9 | 17.9 / 17.6 | 28.7 / 17.7 |
| S6.2 | pink T-15, 10 s room tone at -50, bursts at T | 130 | ‑0.0 | 3.8 / 2.3 | 2.8 / 1.8 | 16.9 / 16.2 | 31.5 / 28.5 |
| S8.-60.30 | 30 s room tone at -60, then pink T | 30 | +0.0 | 7.4 / 4.4 | 4.8 / 1.7 | 23.5 / 26.6 | 23.6 / 91.6 |
| S8.-60.120 | 120 s room tone at -60, then pink T | 120 | +0.0 | 7.4 / 5.0 | 6.0 / 3.8 | 38.6 / 38.7 | 141.1 / 98.2 |
| S8.-60.300 | 300 s room tone at -60, then pink T | 300 | +0.0 | 7.4 / 5.0 | 6.0 / 3.8 | 38.6 / 38.7 | 141.1 / 98.2 |
| S8.-50.30 | 30 s room tone at -50, then pink T | 30 | +0.0 | 7.4 / 4.1 | 4.3 / 3.9 | 18.1 / 22.2 | 88.1 / 37.4 |
| S8.-50.120 | 120 s room tone at -50, then pink T | 120 | +0.0 | 7.4 / 5.4 | 5.9 / 3.9 | 23.6 / 23.8 | 24.1 / 24.5 |
| S8.-50.300 | 300 s room tone at -50, then pink T | 300 | +0.0 | 7.4 / 5.4 | 5.9 / 3.9 | 23.6 / 23.8 | 24.1 / 24.5 |
| S8.-45.30 | 30 s room tone at -45, then pink T | 30 | +0.0 | 7.4 / 4.1 | 4.3 / 3.0 | 16.7 / 14.1 | 82.2 / 26.9 |
| S8.-45.120 | 120 s room tone at -45, then pink T | 120 | +0.0 | 7.4 / 4.3 | 5.5 / 3.0 | 19.8 / 17.7 | 90.1 / 45.4 |
| S8.-45.300 | 300 s room tone at -45, then pink T | 300 | +0.0 | 7.4 / 4.3 | 5.5 / 3.0 | 19.8 / 17.7 | 90.0 / 44.9 |
Top: output short-term loudness, with the ±1 LU band. Bottom: the gain applied, against the gain a perfect fixed gain would apply (dashed). Green: mode 0. Amber: mode 1. White: mode 2, drawn where it differs from mode 1 by more than 1 dB.
S1.+12 A 12 LU step up: too loud for about 40 s, then settled.S1.-12 A 12 LU step down: the lift arrives over about a minute.S2.20 Loud material after a long quiet stretch: the gain arrives on the +12 dB maximum. Mode 2 differs from mode 1 here by up to 4.0 dB.S3.up The same 12 dB step on programme material. Mode 2 differs from mode 1 here by up to 4.5 dB.S4.L.30 30 s of digital silence in material that needs a lift.S5.-50.120 Two minutes of room tone at -50 LUFS in a pause.S6.1 Noise bursts after a quiet passage and 10 s of silence. Mode 2 differs from mode 1 here by up to 2.5 dB.S8.-50.120 The stream starts on two minutes of room tone. Mode 2 differs from mode 1 here by up to 3.0 dB.Sine tones at 40, 63 and 100 Hz that step from 20 LU under the target to 10 LU over it, where the limiter takes the step, and a 50 Hz kick loop. Low-frequency distortion as the method defines it: harmonics 2 to 10 against the fundamental, Hann window. Reported, without a pass line.
| Case | Signal | THD % 20-200 ms after the step, modes 0 / 1 / 2 | THD % 0.3-1.3 s after, modes 0 / 1 / 2 | Loudest momentary over the target LU, modes 0 / 1 / 2 |
|---|---|---|---|---|
| L1.40 | 40 Hz tone T-20 -> T+10 | 0.355 / 0.219 / 0.219 (input 0.051) | 0.408 / 0.001 / 0.001 | 6.21 / 6.79 / 6.79 |
| L1.63 | 63 Hz tone T-20 -> T+10 | 0.070 / 0.183 / 0.183 (input 0.019) | 0.547 / 0.003 / 0.003 | 7.71 / 7.91 / 7.91 |
| L1.100 | 100 Hz tone T-20 -> T+10 | 0.337 / 0.098 / 0.098 (input 0.000) | 0.492 / 0.001 / 0.001 | 8.59 / 8.50 / 8.50 |
Integrated output, LUFS. The highest true peak at -16 LUFS is ‑1.00 dBTP: the limiter holds every kick at the ceiling.
| Target | Mode 0 | Mode 1 | Mode 2 | Fixed-gain two-pass reference |
|---|---|---|---|---|
| ‑23 LUFS | ‑22.23 | ‑23.00 | ‑23.00 | ‑23.00 |
| ‑16 LUFS | ‑15.67 | ‑17.32 | ‑17.32 | ‑17.42 |
Every decision is taken on a 10 ms grid counted from the start of the stream, so the output does not depend on how the host cuts the stream into calls. Checked by a 64-bit hash of every output sample, per 100 ms. The last row of each mode is the library against the tuned lab chain it was ported from.
| Mode | Block size | Cases identical | Differences |
|---|---|---|---|
| 0 | 20 ms | 108 of 108 | every output sample identical |
| 0 | 64 frames | 49 of 49 | every output sample identical |
| 0 | 256 frames | 108 of 108 | every output sample identical |
| 0 | 512 frames | 108 of 108 | every output sample identical |
| 0 | 1024 frames | 108 of 108 | every output sample identical |
| 0 | 2048 frames | 107 of 108 | S7: from 120.0 s, 85 of 3000 ticks |
| 0 | the library against the tuned lab chain, 10 ms | 45 of 45 | every output sample identical |
| 1 | 20 ms | 112 of 112 | every output sample identical |
| 1 | 64 frames | 53 of 53 | every output sample identical |
| 1 | 256 frames | 112 of 112 | every output sample identical |
| 1 | 512 frames | 112 of 112 | every output sample identical |
| 1 | 1024 frames | 112 of 112 | every output sample identical |
| 1 | 2048 frames | 111 of 112 | S7: from 120.0 s, 81 of 3000 ticks |
| 1 | the library against the tuned lab chain, 10 ms | 117 of 117 | every output sample identical |
| 2 | 20 ms | 112 of 112 | every output sample identical |
| 2 | 64 frames | 53 of 53 | every output sample identical |
| 2 | 256 frames | 112 of 112 | every output sample identical |
| 2 | 512 frames | 112 of 112 | every output sample identical |
| 2 | 1024 frames | 112 of 112 | every output sample identical |
| 2 | 2048 frames | 111 of 112 | S7: from 120.0 s, 81 of 3000 ticks |
| 2 | the library against the tuned lab chain, 10 ms | 117 of 117 | every output sample identical |
The call that starts each 10 ms step does the adjuster's periodic work, so it is the slowest. The smallest host blocks at which that call takes at most half of the block, in 999 calls of 1000, single-threaded on an otherwise idle machine, which leaves the host the other half. Measured over 1 h 00 min of wide-ranging music (P10.1); the mean CPU on P9.15.
| Mode | Latency | Minimum block at 48 kHz, up to an hour after Create or Reset | Continuous operation (estimated) | Mean CPU per call, share of the block: 64 frames / 10 ms |
|---|---|---|---|---|
| 0, standard | 50 ms | 64 frames (1.3 ms) | 128 frames | 4.2 % / 3.6 % |
| 1, low latency | 2 ms | 64 frames (1.3 ms) | 128 frames | 4.2 % / 3.6 % |
| 2, low latency with dialogue | 2 ms | 128 frames (2.7 ms) | 256 frames | 5.0 % / 4.3 % |
Frames at 48 kHz; at 44.1 kHz the same frame counts leave more time. The adjuster's long-term meters
cost more as the range of loudness they have heard since Create or Reset()
grows, so a stream that runs for many hours without a reset needs the larger block. That column is
estimated from the measured cost per meter step, not measured over days.
| Mode | 64 frames | 128 frames | 256 frames |
|---|---|---|---|
| 0, standard | 48 % | 23 % | 14 % |
| 1, low latency | 42 % | 22 % | 12 % |
| 2, low latency with dialogue | 72 % | 38 % | 20 % |
The cases longer than 16 minutes, with the default (the whole stream) and with a 900 s program length. Neutral on an hour of material; on three hours and more it keeps the long-term level on target.
| Case | Material | Length | Long-term in band %, mode 0: default / 900 s | Long-term in band %, mode 1: default / 900 s | Long-term in band %, mode 2: default / 900 s | Integrated LUFS at 900 s, modes 0 / 1 / 2 | W300 in band % at 900 s, modes 0 / 1 / 2 |
|---|---|---|---|---|---|---|---|
| P2a | Zapping stream, recordings 1-11 | 54 min 58 s | 99.1 / 99.1 | 99.5 / 99.5 | 99.5 / 99.5 | ‑22.82 / ‑22.82 / ‑22.69 | 99.5 / 99.5 / 96.5 |
| P7 | Poetry, twenty readings in twelve languages | 40 min 23 s | 99.5 / 99.5 | 99.5 / 99.5 | 99.7 / 99.7 | ‑23.07 / ‑23.08 / ‑23.02 | 100.0 / 100.0 / 100.0 |
| P8.2 | Audiobook, The Cat of Maurice | 39 min 50 s | 99.4 / 99.4 | 99.3 / 99.3 | 98.8 / 98.8 | ‑23.17 / ‑23.19 / ‑23.06 | 100.0 / 100.0 / 100.0 |
| P10.1 | Music mix 1, classical | 1 h 00 min | 98.8 / 98.8 | 99.7 / 99.7 | 99.7 / 99.7 | ‑22.79 / ‑22.80 / ‑22.80 | 90.9 / 88.3 / 88.3 |
| P10.2 | Music mix 2, chart pop | 1 h 03 min | 99.7 / 99.7 | 100.0 / 100.0 | 100.0 / 100.0 | ‑22.50 / ‑22.49 / ‑22.49 | 94.2 / 91.9 / 91.9 |
| P10.3 | Music mix 3, pop | 1 h 10 min | 100.0 / 100.0 | 99.8 / 99.8 | 92.5 / 92.5 | ‑22.39 / ‑22.39 / ‑22.38 | 94.1 / 94.2 / 92.0 |
| P10.4 | Music mix 4, rock and pop | 31 min 34 s | 100.0 / 100.0 | 100.0 / 100.0 | 100.0 / 100.0 | ‑22.48 / ‑22.48 / ‑22.48 | 100.0 / 100.0 / 100.0 |
| P10.5 | Music mix 5, workout | 1 h 00 min | 99.5 / 99.5 | 99.6 / 99.6 | 57.6 / 63.0 | ‑22.29 / ‑22.29 / ‑22.05 | 99.6 / 99.6 / 58.6 |
| P11 | Mixed 1, monologue over an intermittent bed | 59 min 33 s | 97.5 / 97.5 | 99.0 / 99.0 | 99.0 / 90.7 | ‑22.75 / ‑22.77 / ‑22.67 | 98.7 / 97.5 / 73.1 |
| P12 | Live stream 1, game audio and voice | 30 min 36 s | 87.0 / 87.0 | 88.1 / 88.1 | 23.9 / 43.6 | ‑22.89 / ‑23.00 / ‑23.56 | 91.0 / 87.4 / 53.5 |
| P13.1 | Conversation 1, European Commission midday briefing (CC BY 4.0) | 49 min 48 s | 99.9 / 99.9 | 99.9 / 99.9 | 100.0 / 100.0 | ‑23.10 / ‑23.12 / ‑23.09 | 100.0 / 100.0 / 100.0 |
| P13.2 | Conversation 2, FOMC press conference (public domain) | 28 min 42 s | 99.9 / 99.9 | 99.9 / 99.9 | 99.9 / 99.9 | ‑23.12 / ‑23.11 / ‑23.10 | 100.0 / 100.0 / 100.0 |
| P13.3 | Conversation 3, panel discussion, TIB AV-Portal (CC BY 3.0 DE) | 1 h 24 min | 96.8 / 96.7 | 91.6 / 91.4 | 93.8 / 93.8 | ‑23.43 / ‑23.53 / ‑23.47 | 85.8 / 80.5 / 84.2 |
| P13.4 | Conversation 4, US Supreme Court oral argument (public domain) | 1 h 49 min | 99.8 / 99.8 | 99.9 / 99.9 | 99.9 / 99.9 | ‑22.68 / ‑22.71 / ‑22.68 | 100.0 / 100.0 / 100.0 |
| P6 | Long run, four mixes back to back (workout, mixed, pop, rock) | 3 h 42 min | 92.2 / 99.9 | 89.4 / 99.9 | 84.3 / 90.1 | ‑22.45 / ‑22.45 / ‑22.36 | 94.5 / 94.9 / 76.9 |
P6 Long-term loudness over 3 h 42 min with a 900 s program length: mode 0, mode 1, mode 2; dashed, mode 0 with the default. The ±1 LU band.The programme cases and the kick loop again at the streaming target. The other synthetic cases stay at -23 LUFS, where their levels are defined.
| Pass line | Scope | Mode 0, standard | Mode 1, low latency | Mode 2, low latency with dialogue |
|---|---|---|---|---|
| G1 | Whole-case integrated output within ±1 LU of ‑16 LUFS | 63 of 67 | 63 of 67 | reported: 59 of 67 |
| G1-D | Output dialogue loudness within ±1 LU, cases with at least 25 % speech | reported: 42 of 45 | reported: 43 of 45 | 43 of 45 |
| G2 | Long-term loudness in band, cases longer than 16 min | 11 of 14 | 11 of 14 | reported: 10 of 14 |
| G3 | Maximum true peak at or under ‑0.8 dBTP on the lossless cases run at this target (the EBU / ITU files and the kick loop) | 5 of 5; highest ‑0.88 dBTP | 5 of 5; highest ‑1.00 dBTP | 5 of 5; highest ‑0.97 dBTP |
| Line | Case | Mode | Material | Measured | What it is |
|---|---|---|---|---|---|
| G1 | P7.02 | 0, 1 | Poetry, French 2 | 0: ‑17.65 LUFS 1: ‑17.49 LUFS | Needs +12.4 dB. Near or beyond the +12 dB the adjuster adds at most, so the target is reached slowly or not at all. |
| G1 | P7.03 | 0, 1 | Poetry, German 1 | 0: ‑17.28 LUFS 1: ‑17.26 LUFS | A 132 s reading that needs a large lift: it ends before the ride has settled. |
| G1 | P9.02 | 0, 1 | Big Buck Bunny (Blender, CC-BY) | 0: ‑17.39 LUFS 1: ‑17.32 LUFS | A quiet film with loud peaks: the true-peak ceiling holds it back, not the boost maximum. |
| G1 | P13.3 | 0, 1 | Conversation 3, panel discussion, TIB AV-Portal (CC BY 3.0 DE) | 0: ‑18.79 LUFS 1: ‑18.76 LUFS | Needs +13.0 dB. Near or beyond the +12 dB the adjuster adds at most, so the target is reached slowly or not at all. |
| G1-D | P7.02 | 2 | Poetry, French 2 | dialogue ‑17.48 LUFS (programme ‑17.50) | Needs +12.4 dB. Near or beyond the +12 dB the adjuster adds at most, so the target is reached slowly or not at all. |
| G1-D | P13.3 | 2 | Conversation 3, panel discussion, TIB AV-Portal (CC BY 3.0 DE) | dialogue ‑18.88 LUFS (programme ‑18.71) | Needs +13.0 dB. Near or beyond the +12 dB the adjuster adds at most, so the target is reached slowly or not at all. |
| G2 | P8.2 | 0, 1 | Audiobook, The Cat of Maurice | 0: 89.73 % in band 1: 90.92 % in band | Needs +11.2 dB. Near or beyond the +12 dB the adjuster adds at most, so the target is reached slowly or not at all. |
| G2 | P12 | 0, 1 | Live stream 1, game audio and voice | 0: 50.16 % in band 1: 68.31 % in band | The sparse live stream again, further under a higher target. |
| G2 | P13.3 | 0, 1 | Conversation 3, panel discussion, TIB AV-Portal (CC BY 3.0 DE) | 0: 0.11 % in band 1: 0.13 % in band | Needs +13.0 dB. Near or beyond the +12 dB the adjuster adds at most, so the target is reached slowly or not at all. |
| Line | Case | Mode | Material | Measured | What it is |
|---|---|---|---|---|---|
| G1 | P4.1 | 2 | ITU BS.1770-2 conformance, stereo | ‑14.37 LUFS (dialogue ‑16.17) | An ITU conformance piece whose dialogue sits under the mix: mode 2 holds the dialogue at the target, so the programme sits above it. |
| G1 | P4.2 | 2 | ITU BS.1770-2 conformance, mono | ‑14.54 LUFS (dialogue ‑16.16) | An ITU conformance piece whose dialogue sits under the mix: mode 2 holds the dialogue at the target, so the programme sits above it. |
| G1 | P7.02 | 2 | Poetry, French 2 | ‑17.50 LUFS (dialogue ‑17.48) | Needs +12.4 dB. Near or beyond the +12 dB the adjuster adds at most, so the target is reached slowly or not at all. |
| G1 | P7.03 | 2 | Poetry, German 1 | ‑17.34 LUFS (dialogue ‑16.97) | A 132 s reading that needs a large lift: it ends before the ride has settled. |
| G1 | P9.02 | 2 | Big Buck Bunny (Blender, CC-BY) | ‑17.32 LUFS (dialogue ‑24.22) | A quiet film with loud peaks: the true-peak ceiling holds it back, not the boost maximum. |
| G1 | P9.08 | 2 | Elephants Dream (Blender, CC-BY) | ‑14.70 LUFS (dialogue ‑16.39) | A film with little speech, where mode 2 blends the programme and the dialogue; the dialogue sits under the mix, so the programme lands above the target. |
| G1 | P12 | 2 | Live stream 1, game audio and voice | ‑17.42 LUFS (dialogue ‑15.80) | A game stream whose voice sits well above the game audio: mode 2 holds the voice at the target, so the programme sits under it. |
| G1 | P13.3 | 2 | Conversation 3, panel discussion, TIB AV-Portal (CC BY 3.0 DE) | ‑18.71 LUFS (dialogue ‑18.88) | Needs +13.0 dB. Near or beyond the +12 dB the adjuster adds at most, so the target is reached slowly or not at all. |
| G1-D | P7.02 | 0, 1 | Poetry, French 2 | 0: dialogue ‑17.59 LUFS (programme ‑17.65) 1: dialogue ‑17.44 LUFS (programme ‑17.49) | Needs +12.4 dB. Near or beyond the +12 dB the adjuster adds at most, so the target is reached slowly or not at all. |
| G1-D | P9.16 | 0 | The Daily Dweebs (Blender, CC-BY) | dialogue ‑17.27 LUFS (programme ‑15.72) | A film whose dialogue sits under its mix: with the programme at the target, the dialogue sits under it. |
| G1-D | P13.3 | 0, 1 | Conversation 3, panel discussion, TIB AV-Portal (CC BY 3.0 DE) | 0: dialogue ‑18.96 LUFS (programme ‑18.79) 1: dialogue ‑18.92 LUFS (programme ‑18.76) | Needs +13.0 dB. Near or beyond the +12 dB the adjuster adds at most, so the target is reached slowly or not at all. |
| G2 | P8.2 | 2 | Audiobook, The Cat of Maurice | 91.48 % in band | Needs +11.2 dB. Near or beyond the +12 dB the adjuster adds at most, so the target is reached slowly or not at all. |
| G2 | P10.5 | 2 | Music mix 5, workout | 71.36 % in band | A music mix in which the speech detector hears speech. Mode 2 then holds that speech at the target; it sits under the mix, so the music is lifted, for tens of minutes. Mode 1 is the mode for music. |
| G2 | P12 | 2 | Live stream 1, game audio and voice | 14.60 % in band | A game stream whose voice sits well above the game audio: mode 2 holds the voice at the target, so the programme sits under it. |
| G2 | P13.3 | 2 | Conversation 3, panel discussion, TIB AV-Portal (CC BY 3.0 DE) | 0.15 % in band | Needs +13.0 dB. Near or beyond the +12 dB the adjuster adds at most, so the target is reached slowly or not at all. |
Everything but the synthetic signals, the tones and the EBU / ITU files came through lossy sources, so it is kept out of the true-peak grading (G3): a codec's own inter-sample overshoot cannot be told apart from the device's.