Very different bitrates from two AC3 files using LAME's VBR V5 preset + PreGain issue

[numaios]

Hello to everybody!

I've been encoding two audio tracks from a DVD:

Track 1: AC3 2_0 256 kbps
Track 2: AC3 2_0 256 kbps

I used BeSweet 1.5 with lame_enc.dll version 3.97 with this command line: -azid( -c normal -L -3db ) -ota( -g max ) -lame( -v --vbr-new -V 5 -b 32 ), for both encodes.

I got this:
Track 1: MP3 joint-stereo at 123 kbps (48 kHz)
Track 2: MP3 joint-stereo at 83 kbps (48 kHz)

Can anybody tell why is this difference in bitrate?

At the beginning of the processing, when finding maximum gain, it suddenly stops when it arrives to minute 8 and starts encoding. Why is this? Because of the downmixing overflows?

Thank you very much.

By the way, is "-v --vbr-new -V 5" the same as "-V5 --vbr-new"? The first one is the command line created by BeSweetGUI and the second one, the recommended by Hydrogenaudio.

[tebasuna51]

I used BeSweet 1.5 with lame_enc.dll version 3.97 with this command line: -azid( -c normal -L -3db ) -ota( -g max ) -lame( -v --vbr-new -V 5 -b 32 ), for both encodes.
-L -3db not needed, don't exists LFE channel

Track 1: MP3 joint-stereo at 123 kbps (48 kHz)
Track 2: MP3 joint-stereo at 83 kbps (48 kHz)

Can anybody tell why is this difference in bitrate?
Maybe lame not need more bitrate for this track (for instance only voice with left = right channel is easy to encode). This is VBR, don't worry.
At the beginning of the processing, when finding maximum gain, it suddenly stops when it arrives to minute 8 and starts encoding. Why is this?
Maybe a 0 dB maximum is reached and don't need analyze more.
Because of the downmixing overflows?
What downmix?, here is 2.0 to stereo not needed downmix.
By the way, is "-v --vbr-new -V 5" the same as "-V5 --vbr-new"? The first one is the command line created by BeSweetGUI and the second one, the recommended by Hydrogenaudio.
The second one is the same, but for Lame.exe (command line encoder for wav/mp3 inputs)

[numaios]

Thanks for your help!

Maybe lame not need more bitrate for this track (for instance only voice with left = right channel is easy to encode)Track 1 is Spanish stereo and track 2 is Catalan stereo, both of them audios for the same movie (Twin Peaks: Fire Walk With Me). Maybe the Catalan track was recorded in poor quality?

This is VBR, don't worry.So I can believe that LAME has encoded track 1 with the same possible quality as track 2?

What downmix?, here is 2.0 to stereo not needed downmix.I know. That surprised me, but in the log file there are several warnings about "downmixing overflow".

The second one is the same, but for Lame.exe (command line encoder for wav/mp3 inputs)So with BeSweet + dll version I always have to use -v --vbr-new -V 5? Lame_enc.dll wouldn't understand -V5 --vbr-new?

[tebasuna51]

Maybe the Catalan track was recorded in poor quality?
Or more acute voices in spanish track, or ...
So I can believe that LAME has encoded track 1 with the same possible quality as track 2?
Yes, this is the idea.
I know. That surprised me, but in the log file there are several warnings about "downmixing overflow".
Can be warnings at encoding phase. Use -ota( -hybridgain ) more recommendable for mp3 output.
So with BeSweet + dll version I always have to use -v --vbr-new -V 5?
Yes.
Lame_enc.dll wouldn't understand -V5 --vbr-new?
You must respect the BeSweet sintax (http://forum.doom9.org/showthread.php?p=629232#post629232), don't worry about how BeSweet send the parameters to lame_enc.dll.

Really there are different behavior between BeSweet-Lame_enc.dll and Lame.exe, like you can see about -b 32 in this post (http://forum.doom9.org/showthread.php?p=806465#post806465)

[numaios]

Can be warnings at encoding phase. Use -ota( -hybridgain ) more recommendable for mp3 output.I tried hybridgain (and AZID's -maximize), and I got warnings too (although in different moments), and that's why I tried PreGain. By the way, isn't PreGain slower but more accurate than HybridGain, even for MP3 output?

don't worry about how BeSweet send the parameters to lame_enc.dllOK, I get it now.

Really there are different behavior between BeSweet-Lame_enc.dll and Lame.exe, like you can see about -b 32 in this postYes, I found out this issue when I was testing VBR mode, and since then I've always added the -b 32 command.

Thank you very much :)

Another question: if LAME decides that 83 kbps are enough, it should do automatic downsampling, shouldn't it?

[tebasuna51]

I tried hybridgain (and AZID's -maximize), and I got warnings too (although in different moments), and that's why I tried PreGain. By the way, isn't PreGain slower but more accurate than HybridGain, even for MP3 output?
A Pre(encoder)Gain 100% is the same than Azid -maximize (PostdecoderGain), but Hybridgain try to avoid encoder overflows with a PreGain less than 100% and a PostEncodeGain to reach the 100%, AFAIK this is more slower and accurate.
Another question: if LAME decides that 83 kbps are enough, it should do automatic downsampling, shouldn't it?
Yes, for this bitrate I get a downsampling from 48000 Hz to 44100 Hz.

[numaios]

Yes, for this bitrate I get a downsampling from 48000 Hz to 44100 Hz.The command line would be -azid( -c normal ) -ota( -hybridgain ) -ssrc( --rate 44100 ) -lame( -v --vbr-new -V 5 -b 32 ) or you get 44100 Hz automatically?

I ask this because, if I remember correctly, I got 83 kbps and 48 kHz with VBR V5 (I'm not sure since I'm not posting from the computer where I encoded it).

[tebasuna51]

Let Lame to do the resample, if needed, automatically.

[numaios]

Let Lame to do the resample, if needed, automatically.OK, I won't worry anymore if I get a 48 kHz/83 kbps file. :)

:thanks: