Help - Search - Members - Calendar
Full Version: Advice for aac encoding speech at <30kbps
Hydrogenaudio Forums > Lossy Audio Compression > AAC > AAC - Tech
Theorist
Please could anyone advise on an aac-lc encoder which can emulate the quality I am getting in mp3 by using the lame settings "-V9 --vbr-new". Lame encodes in stereo and automatically resamples to 24kHz.

I have some speech to encode which has fairly long pauses in it and I want to encode at an average rate of around 28kbps for my ipod. Encoding as mp3 works very well and using the settings "lame -V9 --vbr-new" with version 3.97 gives very good quality (almost transparent!). It can do this because the variable rate drops to 8kbps in the pauses and goes up to 112kbps where necessary. However I want to have the speech in aac format so I can use the bookmarking on my ipod. I have struggled to find an encoder for aac that works anything like as well. Yet everyone seems to suggest that aac should be even better than mp3 at low bit rates.

I have tried the neroencoder, iTunes and aacplusenc. iTunes and nero I find unlistenable at those bitrates on the ipod. Of the three the best I have found is aacplusenc but of course this encodes in HE-AAC+v2 and so on my ipod it plays in mono and with half the frequency range. While the quality is not bad, I would have thought I could at least get similar results to the mp3 encoding. I think the issue is that most of the aac encoders use a relatively limited range around the target bitrate in implementing VBR - but I am no expert. I think I must be using the wrong settings.

Does anyone have any better suggestions of an aac encoder with the appropriate settings? I can use either linux (prefered) or win32.
Thanks
Garf
What settigns did you use for the Nero encoder?

Did you try resampling to audio and then giving it to the encoders?
Theorist
QUOTE(Garf @ Jan 30 2008, 17:53) *

What settigns did you use for the Nero encoder?

Did you try resampling to audio and then giving it to the encoders?


I tried various settings, forcing lc for example, setting a target quality that was low or using -br with two pass. There is not much flexibility from the command line. For example

neroAacEnc -lc -q 0.1 -if tests.wav -of nero.aac

This gets the bitrate about right, but sounds much worse than the slightly more compressed

lame -V9 --vbr-new tests.wav tests.mp3

I did not try resampling the original. I will give that a go and report back...Thanks
benski
Winamp's AAC encoder has an option for speech that work surprisingly well. However, it is CBR so you won't benefit too much from the long periods of silence. Worth a shot, though.
Theorist
Just to follow up with some of the settings I have tried in response to the suggestions so far.
The following settings all give the same filesize
There are two source files sample_44.wav (original 44.1KHz) and sample_22.wav (resampled to 22050Hz)

Best quality:

lame -V9 --vbr-new sample_44.wav lame.mp3

aacplusenc sample_44.wav aacplus44_28.aac 28
MP4Box -add aacplus44_28.aac:sbr aacplus44_28.m4a

(Of course "aacplus44_28.m4a" plays in mono on my ipod and is effectively resampled down to 22kHz too. On a player that can handle HE-AACv2 then I can go down to 20kbps and it still sounds good). However, what I want is to reproduce what I had with lame in lc-aac format for the ipod.)

Quite good:

neroAacEnc -q 0.25 -lc -if sample_22.wav -of nero22_q25.m4a

Fair:

neroAacEnc -lc -br 27000 -2pass -if sample_22.wav -of nero22_2br27.m4a

winamp encoding to LC-AAC at 28kbps speech stereo. It resamples to 11kHz.

Poor:

Anything neroAacEnc does with the sample_44.wav
Garf
You might also want to try resampling to 16kHz and/or 11kHz. Maybe even 32kHz. It will have a strong influence with Nero.
This is a "lo-fi" version of our main content. To view the full version with more information, formatting and images, please click here.
Invision Power Board © 2001-2008 Invision Power Services, Inc.