Here's my (cross posted and slightly altered) reply.

No soundcard suggestions, but one other option to consider perhaps: a USB soundcard.

An advantage: same sound system on every computer you work on, even on laptops in which you can never have the same type of card as in your desktop . This is good if you work with speaker dependant speech recognition, because your soundcard can alter your voice so your acoustic model needs to be adapted to the combination mic+card. So your speech recognition solution becomes more portable when you can take your (USB) soundcard with you.

Disadvantages: might be hard to get to work with some programmes (though this will only come into play under linux I guess), might draw a tiny bit more power on a laptop?

--- (Edited on 8/13/2007 9:46 am [GMT-0500] by Robin) ---

Hi Robin!

Yes, a USB soundcard is a good idea.  But at the moment, I'm using only one desktop computer, no laptop.  And a USB soundcard might be more expensive than a PCI soundcard.

And sure, I would like to use the soundcard under Windows XP *and* under Ubuntu Linux.  My onboard audio works under Windows XP and under Ubuntu Linux, but the recording quality under Windows XP is much better. This is the reason why I am using Windows XP to create the audio files that I submit to VoxForge.org.  

I found a one year old discussion about the USB soundcard "Terratec Aureon 5.1 USB MKII", which might be appropriate for Linux:

http://tinyurl.com/2pqzey (German language)

This USB soundcard should work under Linux -, but at the moment I don't know whether it provides a better recording quality than my onboard audio (Mainboard: AsRock 939Dual-SATA2).

--- (Edited on 8/14/2007 10:52 am [GMT-0500] by ralfherzog) ---

It's quite hard to find a site that gives info on both compatibility and quality. This site: http://bugtrack.alsa-project.org/main/index.php/Matrix:Main should be helpful for compatibility. If a model is supported then check how good it is compared to your current card.

--- (Edited on 8/15/2007 10:50 am [GMT-0500] by Robin) ---

Hi Robin!

Thanks for the tip.  At the moment, I am thinking about buying the sound card "Creative Sound Blaster X-Fi XtremeMusic PCI", but I could see in the alsa-project.org wiki that this sound card is *not* supported (unknown chipset, completely new architecture, manufacturer doesn't support the ALSA developers).

So thanks for the tip, now I know more.

The "Creative Sound Blaster X-Fi XtremeMusic PCI" has the advantage that it is possible to record in the quality of 24-bit/96 kHz.  Would it make sense to record in such a high resolution?  You people from voxforge.org say that you are interested in recordings of the quality 16-bit/48 kHz.  But why not use the higher quality if available?

--- (Edited on 8/16/2007 7:42 pm [GMT-0500] by ralfherzog) ---

Hi Ralf,

>But why not use the higher quality if available?

This formula calculating audio file sizes is as follows: 

File size in bytes = sample rate (Hz) x  bits (8/16/24) x channels (1,2,...) x seconds / 8

Therefore, if we collect mono audio with a 48kHz sampling rate, at 16 bits per sample, our target release 1.0 corpus of 140 hours of speech would be at least 45 Gig.

48,000Hz * 16 bits* 1 channel *  (140hours*60*60)/8 = 45.06 Gbytes

If we collect mono audio with a 96kHz sampling rate, at 16 bits per sample, our target corpus of 140 hours of speech would be about 90 Gig. 

96,000Hz * 16 bits* 1 channel *  (140hours*60*60)/8 = 90.12 Gbytes

This is just for release 1.0 of the VoxForge English Corpus.  Since we are starting to look at other languages, the storage and bandwidth requirements for moving to a higher sampling rate would be too large at this stage.

Ken

--- (Edited on 8/17/2007 2:47 pm [GMT-0400] by kmaclean) ---

--- (Edited on 8/19/2007 9:34 am [GMT-0400] by kmaclean) ---

Hello Ken,

I understand that there are storage and bandwidth requirements.  But what about using the "Free Lossless Audio Codec" instead of WAV?

Take a look at http://en.wikipedia.org/wiki/Free_Lossless_Audio_Codec !

"FLAC's main advantage is the reduction of bandwidth or storage requirements, but without sacrificing the integrity of the audio source. [...] FLAC is for efficient packing of audio data[...] FLAC achieves compression rates of 30 - 50% for most music, with significantly greater compression for voice recordings."

Read the last part again: "significantly greater compression for voice recordings", and think about that!  If the compression rate using FLAC would be more than 50% (at 96.000 Hz/16-bit) then we wouldn't need more storage and bandwidth.

I have no experience with FLAC, but this could be worth a thought.  What is your opinion?  Keep in mind: FLAC is *lossless* compression.

Greetings, Ralf

--- (Edited on 8/17/2007 5:18 pm [GMT-0500] by ralfherzog) ---

Hi Ralph,

Thanks for pointing out FLAC lossless compression as an alternative to uncompressed WAV, AIFF or raw.

We actually accept LibriVox submissions in FLAC format (no takers yet ...) - see this FAQ.  And the second iteration of the new speech submission app will likely use FLAC rather than WAV, for the reasons you've mentioned.

As an aside, I've not tested FLAC, so I don't know how valid their claims of a 50% reduction in file is, but even a 25% reduction would make it worthwhile.

Ken 

--- (Edited on 8/18/2007 3:23 pm [GMT-0400] by kmaclean) ---

Hello Ken,

Because you accept submissions in the FLAC format, I will try to convert my wav audio files into the FLAC format before zipping them.

A few minutes ago, I have installed a FLAC converter from the following source: http://tinyurl.com/33upwp

I made a test and converted a wav speech file with the size of 400 kB into a FLAC file.  The result was the size of only 140 kB.

Greetings, Ralf

--- (Edited on 8/18/2007 9:49 pm [GMT-0500] by ralfherzog) ---

--- (Edited on 8/18/2007 9:53 pm [GMT-0500] by ralfherzog) ---

Apparently you can even export your sound files directly as FLAC from Audacity (depending on your installation and version).

See this link: http://audacityteam.org/wiki/index.php?title=FLAC

Robin 

--- (Edited on 8/21/2007 10:28 am [GMT-0500] by Robin) ---