whether to use VBR or CBR when re-encoding audio

[chronis]

When your source audio has CBR below the maximum possible bitrate (e.g. AC3 192kbps) and you need to re-encode it with almost no loss in quality, is it better to use VBR instead of CBR as the output?

For example, AC3 has maximum bitrate of 640kbps, so if my source is AC3 640kbps CBR, then I'd want to choose 640 CBR as the output to retain the highest quality, right?

However, if I'm converting AC3 at 192 CBR, would it be better to use something like AAC 192 VBR? (since AC3 only comes in CBR)
Or is using VBR in this case pointless since no part of the audio has data exceeding 192kbps?

(Note: I know that re-encoding lossy audio is not recommended, but this is in cases where you have to, for example when converting 25fps to 24fps.)

[tebasuna51]

Audio CBR is only needed to make DVD's or AVI's, if you want use modern containers (mp4, mkv) and players always use VBR if you need recode.

My recomendation is use qaac TrueVBR -V 91 for sources like a stereo AC3 192 Kb/s, the Constrained mode is only if you want control the container size.
We don't know if the source CBR bitrate was well adjusted to the audio quality, maybe some parts need the full 192 Kb/s in AC3 but of course less in AAC.
The AAC encoder can have troubles to obtain a CVBR 192 Kb/s from a source 192 Kb/s AC3 CBR, not need so much bitrate.

[chronis]

My recomendation is use qaac TrueVBR -V 91 for sources like a stereo AC3 192 Kb/s, the Constrained mode is only if you want control the container size.
We don't know if the source CBR bitrate was well adjusted to the audio quality, maybe some parts need the full 192 Kb/s in AC3 but of course less in AAC.
The AAC encoder can have troubles to obtain a CVBR 192 Kb/s from a source 192 Kb/s AC3 CBR, not need so much bitrate.

Thanks tebasuna51, this is very helpful. The only thing I'm still confused about is this: if the audio bitrate is constant, never exceeding 192kb/s, and we want to maintain the highest quality by keeping the bitrate at around 192, why is it preferable to re-encode to VBR instead of CBR? Doing so would mean that some parts of the audio exceed 192 (since it's variable), yet the source audio never exceeded 192 to begin with. So why is it preferable?

[tebasuna51]

I recode a 43 minutes track AC3 192 Kb/s using qaac -V 91 and obtain a file of 44,8 MB with:

TVBR min./avg./max. : 6 / 145 / 461 Kb/s.

Using qaac -v 192 the file size is 61,4 MB with:

CVBR min./avg./max. : 6 / 199 / 468 Kb/s.

Do you think than the 61,4 MB is better than the 44,8 MB ?

From the same source a AAC CBR 145 Kb/s is better than the 192 Kb/s AC3
With the CVBR we waste bitrate encoding transients, lossy encoder/decoder errors, noise ... nothing from the initial original source.

[chronis]

@tebasuna51 Allow me to clarify my question with an example. The TVBR file you obtained has a maximum bitrate of 461kb/s, however the source had a CBR of 192kb/s, which by definition is also the maximum bitrate. Isn't it in a sense pointless to have bitrates exceeding 192? This is what I'm trying to understand, if you can explain.

I want to retain the highest quality possible, regardless of size. Can you confirm that re-encoding 192 CBR-->192 VBR gives you better quality than 192 CBR-->192 CBR? (I'm using 192 as an example)

[tebasuna51]

...the TVBR file you obtained has a maximum bitrate of 461kb/s, however the source had a CBR of 192kb/s, which by definition is also the maximum bitrate. Isn't it in a sense pointless to have bitrates exceeding 192?...

You need understand the lossy encode/decode process:

1) The original source is a analog audio recorded with analog mics than can introduce noise.

2) That source is digitalized with more or less precission (16, 24, 32... bits) but always introducing some quantization noise.

3) The first encoder (maybe AC3 at 192 Kb/s) convert the samples in time domain in samples in frequency domain using float numbers, and select the most important ones (by volume and frequency) than fit in the bitrate assigned, introducing errors.

4) The decoder convert samples in frequency domain in samples float in time domain, now downconvert (if requested) to int samples introducing round noise.

If you timestretch that audio you introduce more errors, and you finish with an audio than can need much more bitrate to be recode with precission.
But, for what you need encode with precission all errors and noise introduced in the different steps?
High bitrate is needed for high frequencies, are there in the original first encode or are they noise?

Don't worry about all errors and high frequencies introduced, most of them are inaudibles, if you select enough quality you can't difference the AC3 192 Kb/s than the AAC -V 91 with a standard audio equipment and ears.

Of course the best option is remain the audio untouched, but if you need recode select a quality according to the original bitrate, maybe for a lossless original encode you can use -V 99, or for a source MP3 128 Kb/s use -V 82 or less.

[tebasuna51]

...Can you confirm that re-encoding 192 CBR-->192 VBR gives you better quality than 192 CBR-->192 CBR?
Of course.

And with movie tracks (silences, dialogs, ...) most the times re-encoding 192 CBR-->TVBR -V 91 gives you better quality than 192 CBR-->192 CBR

[chronis]

@tebasuna51 Thanks for your detailed explanations. I appreciate it.

[chronis]

maybe for a lossless original encode you can use -V 99, or for a source MP3 128 Kb/s use -V 82 or less.

I noticed that the QAAC options for quality go from 0 to 127, with increments of 9, so 91 (default), 100, 109, 118, 127. Wouldn't you say that the closest to lossless is 127?

Also, I did a test with 192kbps CBR audio, and here are the bitrates (output) for 3 different settings:
QAAC 91 - 110kbps
QAAC 100 - 125kbps
QAAC 127 - 168kbps

I'm going to probably stick with 127, even though I know it's inefficient at the highest setting. When you have 10+GB movies with shitty audio, an extra 30MB is meaningless. :D

[tebasuna51]

Test over multiple stereo music in HydrogenAudio

HydrogenAudio test
----------------------
V 0 a 4 = ~ 40 kb/s
V 5 a 13 = ~ 45 kb/s
V 14 a 22 = ~ 75 kb/s
V 23 a 31 = ~ 80 kb/s
V 32 a 40 = ~ 95 kb/s
V 41 a 49 = ~105 kb/s
V 50 a 58 = ~115 kb/s
V 59 a 68 = ~135 kb/s
V 69 a 77 = ~150 kb/s
V 78 a 86 = ~165 kb/s
V 87 a 95 = ~195 kb/s
V 96 a 104 = ~225 kb/s
V105 a 113 = ~255 kb/s
V114 a 122 = ~285 kb/s
V123 a 127 = ~320 kb/s

When is necesary V 91 can output a average of 195 Kb/s

In audio movie tracks I'm not interested in "lossless", I'm interested in "transparent" (when human ears can't distinguise from lossless)

[chronis]

I completely agree with you that we should aim for transparency, not lossless. However, when you share your files with other people who don't understand such concepts, the perception of quality is also important. They might look at the conversion of 192 CBR to 128 TVBR (QAAC 100) as a significant drop in quality and I don't want to have to explain myself :D. So, in my case, it's preferable to keep the final bitrate as close to the original as possible.

[chronis]

@tebasuna51 What QAAC setting would you recommend for re-encoding untouched 5.1 channel AC3 448kbps audio from DVD? And are there any issues with players supporting multi-channel AAC? I don't think I've ever seen 5.1 in AAC format.

[tebasuna51]

@tebasuna51 What QAAC setting would you recommend for re-encoding untouched 5.1 channel AC3 448kbps audio from DVD?
this is the average bitrate in my test over a limited number of 5.1 tracks (from lossless sources, better than DVD tracks):
Qaac 5.1-48KHz
------ ---------
-V 1 119 Kb/s
-V 9 136 Kb/s
-V 18 158 Kb/s
-V 27 177 Kb/s
-V 36 198 Kb/s
-V 45 240 Kb/s
-V 54 277 Kb/s
-V 64 314 Kb/s
-V 73 351 Kb/s
-V 82 385 Kb/s
-V 91 450 Kb/s
-V 99 507 Kb/s
-V 109 581 Kb/s
-V 118 638 Kb/s
-V 125 704 Kb/s
Then -V 91 seems more than enough and the -V 82 seems better than 448 Kb/s AC3, based in that test (https://tech.ebu.ch/publications/tech3324)

And are there any issues with players supporting multi-channel AAC? I don't think I've ever seen 5.1 in AAC format.

This is the problem, the player must decode the AAC and send by HDMI the decoded 5.1 in PCM to the amplifier/receiver 5.1.

Work if you play from PC (with mpc-hc for instance) also with some standalone players (you must verify) but if you play directly with TV it is not standard (many TV players can output only AC3 by SPDIF to receiver)

[hello_hello]

@tebasuna51 What QAAC setting would you recommend for re-encoding untouched 5.1 channel AC3 448kbps audio from DVD? And are there any issues with players supporting multi-channel AAC? I don't think I've ever seen 5.1 in AAC format.

I use QAAC's v91 VBR option most of the time (I think it's the default), but I don't bother re-encoding AC3 as there's not enough reduction in file size/bitrate to make it worthwhile. If you're downmiximg to stereo before re-encoding it's a different story, and of course if you don't want AC3 for a particular reason that's a different story too. Re-encoding DTS with v91 can definitely reduce the file size, often by as much as 1GB for a movie if it's high bitrate DTS (1509 kbps).

Both the DVD players here (via their USB inputs) and both the TV's USB media players happily decode 5.1ch AAC, but I don't have to worry about sending 5.1ch audio via SPDIF to receiver and I mostly use my PC for video playback anyway.

[tebasuna51]

...and both the TV's USB media players happily decode 5.1ch AAC, but I don't have to worry about sending 5.1ch audio via SPDIF to receiver...

I can't understand, how send decoded "5.1ch audio via SPDIF to receiver"?

The SPDIF conexion only allow decoded PCM 2.0, PCM 5.1 need a HDMI conexion.

[hello_hello]

I can't understand, how send decoded "5.1ch audio via SPDIF to receiver"?

The SPDIF conexion only allow decoded PCM 2.0, PCM 5.1 need a HDMI conexion.

Sorry, that's what I meant. You can't do it, but it's not something I need to do. I guess I worded it badly.