Saving 24 bit Wav files?

I’m transferring quite a few reels onto hard drive and Audacity is invaluable, but I do have one question. I’m recording them as 32 bit float and saving them as 24 bit wav. Since there is no option for 24 bit wav, I select other uncompressed files and then Microsoft Wav and signed 24 bit PCM. It works fine but I figured it would be better to use FLAC. When I tried exporting as FLAC (24 bit) it works fine but when I decompress the Flac the resulting wav is maybe 1/3 smaller than if I did the WAV output. I can’t tell any difference soundwise and file properties of the 2 files are identical (24 bit, 2304 kbps, 48 kHz, PCM). So I figure exporting directly to FLAC is find but what is happening with the files I’m saving directly as WAV that causes them to be bigger.

For instance, a 20 minute file directly to wav is 476 mb whereas sending to Flac and uncompressing yields a 348 mb file. Thanks.

How are you converting the FLAC files back to WAV? Are you sure that you are converting from FLAC to 24 bit WAV and not to 16 bit WAV?

20 minutes for 24 bit, 44100 Hz stereo should be about 302 MB.
20 minutes for 24 bit, 48000 Hz stereo should be about 330 MB.
20 minutes for 24 bit, 96000 Hz stereo should be about 660 MB.

The FLAC file should be about 40% smaller than the WAV file (depending on the material).

I just took the FLAC file and ran it through the old FLAC front end and hit Decode. When I look at the properties of the files (WInXP, right click on file select Properties and look at Summary tab) shows that both are 24 bit and 48kHz. Yet there is a significant size difference. The actual length is 20:38. Somehow selecting a direct output into wav of 24 bit 48 kHz just creates a much bigger file than if I go through FLAC.

I can’t reproduce any such thing using the Windows FLAC installer from .

That’s a 32-bit 48000 Hz stereo WAV (about 22 MB/minute). Perhaps XP doesn’t understand 32-bit files.

Try looking at the files in MediaInfo (get the version without installer) .


20 minutes and 39 seconds, stereo, 48kHz sample rate 24 bits will be 356544000 bytes. It doesn’t matter where it comes from or how it got there, that is the size. There may be a few extra bytes due to metadata but nowhere near enough to account for the difference that you are describing.
How have you checked that both of the files are both the same duration?

As a test:

  1. Create a new stereo track (Tracks menu)
  2. Generate a 20 minute tune (Generate menu)
  3. Check that the Project rate is set to 48000 (bottom left corner of the main Audacity interface).
  4. Export as a “signed 24 bit WAV file”.

The file size should be very close to 345600000 bytes (about 329.6 MB).

345,600,044. I don’t have an export option for “signed 24 bit WAV File”. I have to used the Other Uncompressed option.

Well I couldn’t duplicate my issue either so I must have messed it up somehow. But I used MediaInfo to look at the Flac and Wav file and while both are 48/24/2 channel/PCM, the wav shows as 2304 kbps while the flac is 1714. I guess it doesn’t matter because it’s 2304 when it expands. Ok, I’m happy. Thanks.

Yes that’s the correct way to export as 24 bit WAV

I’m recording them as 32 bit float…

FYI - You can’t really “capture” or record 32-bit data… There are no 32-bit or floating-point ADCs (analog to digital converters). There’s no advantage in asking your driver to generate a floating-point file. You just end-up with a bigger file. If your soundcard/interface has a 24-bit ADC, you can record in 24-bit integer format. If you have a 16-bit soundcard/interface, you are only wasting space with 24 or 32 bits. But, 16-bits is OK too… 16-bits gives you more resolution and more dynamic range than anything analog. :wink: (There are no floating-point DACs either.)

Audacity (and most other audio editors) work internally at 32-bit floating-point. So, in some rare special situations it may be worthwhile to save in 32-bit floating point temporarily if you are in the middle of editing/mixing/processing, etc. i.e. if you’ve done some processing to push your peaks over 0dB, etc., saving in integer format will clip (distort) your data, so floating point would be prefered (if you don’t want to reduce your levels before saving). But, there’s no reason to record in floating point or to play-back in floating-point.