[Task] Extend the list of supported audio features

[fabiocat93]

Description

We want to expand the audio features that senselab can extract. This will include:

• Time-varying features: Features that provide detailed information about audio properties over time.
• Aggregated features: Statistical or summarized representations of audio properties (e.g., min, max, range, mean pitch, standard deviation of feature X).

Tasks

• Research and shortlist additional audio features (time-varying and aggregated).
• Define how these features will be implemented in senselab.
• Update senselab's feature extraction API to include the new features.
• Write tests to ensure the correctness and reliability of the new features.
• Update documentation to reflect the changes.

[fabiocat93]

Let's start by using this space to discuss features of interest.

[fabiocat93]

about time-varing feats: @satra (from a slack conv) suggested:

• formants
• pitch
• intensity
• loudness
• glottal flow (derivative of)
• vad (this should detect any vocalizations)
• non-vocal fold source (aspiration/frication)

@fabiocat93 added (also inspired by @Rahul-Brito 's slides - of course, @Rahul-Brito feel free to add and comment further)

• Energy
• Perceptual Linear Predictive (PLP)
• Linear Predictive Coding (LPC)
• Line Spectral Pairs (LSP)
• Spectral Shape descriptors (Spectral centroid, Spectral bandwidth, ...)
• Harmonics-to-Noise Ratio (and maybe speech to noise ration - here is an interesting model for estimating it: https://huggingface.co/pyannote/brouhaha)

[fabiocat93]

here are some good resources that i have filtered and may be helpful since they come with their own implementations:

• https://github.com/Shahabks/my-voice-analysis
• https://github.com/Shahabks/myprosody
• https://osf.io/3wr6k/files/osfstorage
• https://voice-lab.github.io/VoiceLab/#documentation-and-api-reference
• https://osf.io/huz7d/
• https://github.com/drfeinberg/PraatScripts/blob/master/Measure%20Pitch%20and%20HNR.ipynb
• https://github.com/drfeinberg/PraatScripts/blob/master/Measure%20Pitch%2C%20HNR%2C%20Jitter%2C%20Shimmer%2C%20and%20Formants.ipynb
• https://github.com/drfeinberg/PraatScripts/blob/master/Measure%20Pitch%2C%20HNR%2C%20Jitter%2C%20and%20Shimmer.ipynb
• https://github.com/drfeinberg/PraatScripts/blob/master/spectral_tilt.py
• https://github.com/drfeinberg/PraatScripts/blob/master/syllable_nuclei.py

[fabiocat93]

also, it may be cool to have some functions for plotting (parselmouth) feats like in here

[fabiocat93]

Also, @skirdey open-sourced some code for audio metrics calculation of PESQ, CSIG, CBAK, COVL, STOI. It may be interesting to make those feats part of senselab (some we already have).

[satra]

audio metrics calculation of PESQ, CSIG, CBAK, COVL, STOI

pesq and stoi are already computed by torchaudio_squim. perhaps add the others to that section/workflow.