Does anybody know of a software, or a method that can be used analyze a piece of audio to determine the "authentic" bit depth? Basically it would tell you if an audio file that is labeled as 24-bit is in fact a 24-bit file, or if it's actually a 16-bit recording/file that has been encoded in 24-bit format. I have more than one situation currently requiring this solution and I have no idea how to accomplish it. I don't need a lecture on how the difference between 16-bit and 24-bit format is not substantial. Any ideas on how to begin solving this problem are welcomed and appreciated.

I did indeed attempt to find out the "authentic" bit-depth information through different a means for each scenario and I have not been able to obtain the answers at this time. This has left me wondering if there is some sort of method, such as a software, that can determine the information that I am seeking.

 

When you're on OSX or linux you can use the terminal command `file`, which actually looks up the exact bit depth. For RIFF WAVE it for instance uses the following line to read out the BitsPerSample "fmt" subchunk; in this case the little endian short (2 bytes long) value located at byte 34 after the header:

`>>>34 leshort >0 \b, %d bit`

If `file` doesn't support the format you're examining, you're on your own but not at all helpless. Just follow the same path file does manually:

1. Lookup the specification of the format (like http://soundfile.sapp.org/doc/WaveFormat/ for RIFF/WAVE)
2. Open the audio file in a hex editor (like `bless` in Linux, or 'Hex Fiend' in OSX)
3. Using the specification, let the editor search for the header field (like the string 'RIFF' for RIFF/WAVE)
4. In the hex editor move your cursor n-bytes further, where 'n' represents the number of bytes where the wanted header is located
5. In the specification, lookup the length and type of the value (eg. BitsPerSample) you want, select the number of bytes in the hex editor and look at the decimal presentation of the value found, eg. the hex values `0x2`, `0x4`, `0x8`, `0x10`, `0x18` or `0x20`, which translate to the decimal values of : 2, 4, 8, 16, 24 and 32 (bits per sample).

BTW: I've just extended upstream `file` to display details for AIFF files. Since those changes won't be available in your OS anytime soon, look at the changes to see where to look in your hex editor. You want the big endian short (=2 bytes long) sampleSize value located 26 bytes after the string 'COMM' (eg. the so called STREAMINFO header).

Good luck,
Ronald

 

Here is a great little program for that exact task and best of all its free.
http://spek.cc/
Spek (IPA: /spɛk/, ‘bacon’ in Dutch) helps to analyse your audio files by showing their spectrogram. Spek is free software available for Unix, Windows and Mac OS X.
Features

• Supports all popular lossy and lossless audio file formats thanks to the FFmpeg libraries.
• Ultra-fast signal processing, uses multiple threads to further speed up the analysis.
• Shows the codec name and the audio signal parameters.
• Allows to save the spectrogram as an image file.
• Drag-and-drop support; associates with common audio file formats.
• Auto-fitting time, frequency and spectral density rulers.
• Adjustable spectral density range.
• Translated into 19 languages.
  
  

Whenever i have been in doubt of a file i use this and it shows you in the graphic aswell as the writing at the top, you just drop the file onto the program and that's it. As you can see in the pictures above the difference between a Flac file and a 320 Mp3, if you put say a 128 Mp3 file you will see it only rise to just under 16k max so if you have a file that says its higher but shows a cut off at 16k, you know its BS (see below picture for an example).

 

Last edited: Sep 18, 2017

looks like they didnt understand your question. By the way it is cute to see people thinking that it is easy on osx but apparently inexistant on windows. So, the "great secret line of command of linux" BS told here looks intriguining, but in fact just tells you if its 24bits. As far as i know, the whole tingeling done, is easely done by "showing the bitdepth" on your audio-player, like foobar, properties .... oh me oh my, it says the bitdephth ! No need to write command lines ?

I don't think you can find some sort of bullet proof technique or software that can tell or prove you if a file is a 16bits upgraded to 24bits.

 

Stillwell Bitter maybe?

 

In case anyone needs this some time:
You will get the most precise answer from this command:
sox <filename.wav> -n stats
It will calculate statistics and tell you the actual bit depth used.

If the file has been converted from 16bit to 24bit using dither, however, there's practically no way to determine that.

Get sox here

 

No, there is not. Once a file is a 24-bit container that's that. I can record my voice in my daw (as 32-bit), aply an 8-bit bitcrusher and export it as default 24-bit. What's the original? The 8-bit processing? The 32-bit recording? As you can see there's no way to find out.

Edit: Maybe, if someone renders a 24-bit as a 16-bit without dithering, you would be able to see something on a spectral analyzer, make yourself such files and see if you can spot differences.

 

a 16 bit file has 96db headroom, if it has been saved as 24bit you should be albe to tell boosting a silent area by this amount , if theres still silence its prolly a real 24 bit file

 

Just play it through Stillwell Bitter free VST plugin and see how many bits it is. That's the most reliable way. There's also a free SSL plugin X-ISM that tells you if there are ISPs in your file and it's got a bit depth meter, too. http://www.sonicstate.com/news/2007/10/30/free-ssl-plug-in/

 

Finally someone actually thinking about what bit depth really is. It's about noise floor guys!

Of course, that will only tell you definitively if it was 24-bit, but you can't really definitely say something isn't 24-bit because who knows what came before in the signal chain to dirty up the signal (as @jayxflash was alluding to)?

It's one of those "you can prove aliens exists, but you can't prove aliens DON'T exist" kind of things.

 

Last edited: Sep 18, 2017

That's great for sample rates and filtering issues but bit-depth isn't the same thing.

 

If you're analyzing HR Web purchases, rest assured that 96/24 is 95% upscaled from 44/16 or so, according to my research.

1 is NF analysis.
2 simple DSP such as volume change by any amount on 16 bit file in 24bit DSP, will reflow all the bits
3 Anything recorded is not worth >20bit due to thermal noise. Any volume increases or decreases thereafer will reduce/increase the bit-worth,

 

Yes. And that's the problem. How can you measure noise floor when the signal level is way above noise? Only few commercial samples are long enough to contain "silent" areas.

Exactly. And not every 24bit audio i/f has better signal-to-noise characteristics than a good 16bit interface, but that's a different story.
>16bit is indeed great if your signal is very low-level or if you have a recording with rather high dynamics, then you will definitely hear a difference.

I really absolutely don't care about numbers, as long as the samples sound good. Really.
I have seen so many good 16bit and so many bad 24bit (or 32bit!!) sample libraries...
To save space and DFD bandwidth, I usually even downsample+dither my (normalized) samples to 16bit after processing them because the VST and VSTi and DAWs we use mostly work with 32bit (or higher) word length DSP internally.