API breaking changes summary

[ZZZZzzzzac]

If you happen to find this repo like me and encounter errors when running noise.py, it's because some dependency libraries have changed their APIs, breaking this program. Here are some changes I made to fix that:

• Install sox if you haven't already: https://sourceforge.net/projects/sox/ (common for Windows users like me).
• Add sox to PATH or edit pysndfx.dsp.py, adding cwd=your/sox/path to Popen in the AudioEffectsChain.__call__() method.
• In the functions reduce_noise_centroid_s and reduce_noise_centroid_mb, threshold_l/h must be non-zero, as sox no longer accepts a 0 cutoff frequency for the low-shelf filter. I prefer threshold_l = np.max([np.min(cent), 20]) and threshold_h = np.min([np.max(cent), 20000]), for the common human voice range of 20-20kHz.
• In the function output_file, librosa.output is deprecated. Use sp.io.wavfile.write(destination, sr, y) instead.
• In the function trim_silence, librosa.get_duration now requires only keyword arguments like librosa.get_duration(y=y). Here, I recommend adding the sr keyword for any audio other than the default 22050 sample rate.
• I also recommend adding sr=None to librosa.load in the read_file function to keep the native sample rate, resampling explicitly when you need another sample rate.

Dependencies version for me:

• librosa 0.10.1
• pysndfx 0.3.6
• numpy 1.26.3
• python_speech_features 0.6
• scipy 1.11.4