In MP3, the bits per sample can be 8-bit or 16-bit (right? )

Where is this piece of information located in an MP3 file?

1) no
2) nowhere

The easy explanation: MP3 works totally different and doesn't care about what the bits/sample resolution of your source was.

Cheers!
SG

Ok, what's the non-easy explanation? How can MP3 "not care". Doesn't the ADC care? Surely a resolution exists... Surely it's embedded in the mp3 somewhere...

It uses floating point, not integer numbers, and the output samples are merely floating point approximations of the original input samples. The output resolution is literally as high as the numerical precision of your computer, encoder and decoder will allow. Of course, this resolution is mostly just resolving quantization noise and the like.



The integer resolution exists until its fed into the MP3 encoder, which operates on fp numbers. The decoder then converts these back to integer at whatever resolution you tell it to (8, 16, 24, 32, 64, etc). Obviously theres little point in going that high, but there no reason you can't convert those 64 bit fp numbers into whatever precision you like.

Thanks a lot for the reply.

1. So is the recommendation for decoding to just decode at the highest possible resolution permitted by your decoder (or is there some standard for the fp<-->integer conversion)?

2. Where is the fp number stored in the MP3 file?

3. Does anyone know of a good resource for reading about bits per sample on any/all other codecs?

Thanks!

1) Use whatever your soundcard supports. No need to use 64 bits when your soundcard supports only 24 bits.

2) Nowhere.

there are no "bits per sample" in MP3 and other psychacoustic lossy encoders.
In MP3 each of the 32 subbands (not all of them are in use!) allocates its individual amount of bits in each frame

It doesn't matter.



The first couple hundred bytes are the header. The last couple hundred bytes are the file tag, the couple MB in between are your fp numbers.



All lossy formats work like this.

And the fp numbers are always 64 bits each? Are they typically this big for other lossy formats?

Thanks again!




Thanks for the reply.

What is an MP3 subband?

The precision is up to the decoder. Internally, I believe mp3 applies some kind of quantization and then huffman coding. The decoder then decodes this and requantizes back to whatever precision it likes. Typically this will be done using 32 bit fixed or floating point.



http://en.wikipedia.org/wiki/Subband_encoding

Sorry, couldn't resist

Very funny.

Thanks for helpin' out the noob Mike. Much appreciated.