Normalization

[sevagh]

Hello,

I'm using libnyquist for a pitch tracking application. I decode the audio from a wav file, and run it through some algorithms (Mcleod, YIN, etc.)

For a few minutes I was scratching my head about why I was getting completely wrong results until I noticed the normalized between [-1.0, 1.0] part of the README. When I multiply the samples vector output by the decoder by a constant e.g. 100 or 1000 to amplify the signal, my pitch tracking algorithms produce correct results.

My question is,

• Is my approach correct? This decoder library outputting normalized audio data and clients multiplying the vector by a scalar factor to amplify it? Does that introduce any problems, is there a better way for me to achieve that? Modify my pitch tracking code to operate on normalized values?
• Could that behavior be modified in this library? Or is normalized output always desired/mandatory

[ddiakopoulos]

Hi @sevagh - I think that must be a bug in your pitch detection pipeline! Normalized floating point values are quite frequent in audio systems, and many audio APIs will hard clip values over the -1 to 1 range. I'm closing this issue for now but feel free to chime in again if you uncover any mysteries in your code :)