jsingh811/pyaudioprocessing

Audio feature extraction and classification

pyaudioprocessing's Issues

Optimize gfcc feature calculations

Description

The gfcc feature can be optimized in terms of computations and scale. This ticket entails work for the betterment of gfcc feature calculation.

Acceptance Criteria

• GFCC feature improvement
• Separates out chroma and spectral features into individual features rather than features included with gfcc and mfcc

Audio cleaning and visualization

Add audio cleaning methods to remove low-activity regions.
Add audio visualization wrappers.

Add visualization for classifier confusion matrix and cross-validation results

Description

Currently, classifier stats for different hyper-parameters and the final confusion matrix prints to the screen. We want to add functionality that also saves the output to disc and displays it is easy to understand visualizations(ideas welcome!).

Acceptance criteria

-[ ] Shortlist visualization tools
-[ ] Implement and integrate the tool selected after review

Add python 3.6 to workflow

Due to actions/setup-python#544, workflows for Python 3.6 aren't working at this time on latest Ubuntu versions. Once there is a resolution, add Python 3.6 to the workflows.

Add tests for methods within scripts of `features`

Description

Add a pytest script to test functions inside features dir.

Acceptance Criteria

-[ ] Add unit tests for functions in features

Add audio file conversion functionality

The ability to convert audio files from mp4, mp3, mp4 or aac to wav.

Add tests for run_classifications.py

Description

Add pytest script to test functions written in run_classifications.py.

Acceptance Criteria

-[ ] Add unit tests for functions in run_classifications.py

Add mel-spectrogram feature

##Description
A mel spectrogram is a spectrogram where the frequencies are converted to the mel scale.
The images of mel-spectrograms are widely used with CNNs to classify audio.

Is there a problem with the handling of probabilities?

There are two classes in the model, if one has a score of 0.1, surely the other has a score of 0.9? How to modify it to judge the classification one by one?

Write a method that returns mfcc and gfcc features for an input audio file

Currently, mfcc and gfcc features are being computed as a mid step of a classification problem and we don't get a direct view of the features.
One can want to look at these features independently and use them with a different classification backend.

• implement a function to return mfcc for an input audio file
• implement a function to return gfcc for an input audio file
• update documentation to guide the usage of these new functions
  Thus, we want to write a function that outputs mfcc and a function that outputs gfcc feature for an input audio file.

Typos discovered by codespell

% codespell --ignore-words-list="fo" --quiet-level=2

./pyAudioProcessing/extract_features.py:74: stucture ==> structure
./pyAudioProcessing/trainer/audioTrainTest.py:49: containt ==> contain, content
./pyAudioProcessing/trainer/audioTrainTest.py:68: containt ==> contain, content
./pyAudioProcessing/trainer/audioTrainTest.py:242: signle ==> single
./pyAudioProcessing/trainer/audioTrainTest.py:242: seperate ==> separate
./pyAudioProcessing/features/audioFeatureExtraction.py:255: seperate ==> separate
./pyAudioProcessing/features/filters.py:122: eigth ==> eighth, eight
./pyAudioProcessing/features/filters.py:123: eigth ==> eighth, eight

Dependancies need to be updated for Python 3.8 and 3.9

% python3.9 -m pip install --upgrade pip
% python3.9 -m pip install -r requirements/requirements.txt

Will fail because the new pip dependency resolver says the specified version of scipy is not Python 3.9 compatible.

Example dont work

beatExtraction() is used twice but is never defined

Undefined name discovered by #3 https://github.com/jsingh811/pyAudioProcessing/search?q=beatExtraction

install not working on linux

Install not working on linux - unfulfilled dependency python-magic-bin==0.4.14

detail: Centos 8
Clone repo git clone https://github.com/jsingh811/pyAudioProcessing.git

install with pip3 install -e .

Failed dependency

  Downloading https://files.pythonhosted.org/packages/79/db/eaf620b73a1eec3c8c6f8f5b0b236a50f9da88ad57802154b7ba7664d0b8/pydub-0.23.1-py2.py3-none-any.whl
Collecting python-magic-bin==0.4.14 (from pyAudioProcessing==1.1.6)
  Could not find a version that satisfies the requirement python-magic-bin==0.4.14 (from pyAudioProcessing==1.1.6) (from versions: )
No matching distribution found for python-magic-bin==0.4.14 (from pyAudioProcessing==1.1.6)

Reviewing python-magic-bin, I don't see a linux version.

Can you advise?

Write a script to return features for audio files

Currently we can't isolate features from classifications. This feature attempts to isolate the two processes so each can be performed independent of the other.

Add tests for methods within scripts of `trainer`

Description

Add a pytest script to test functions inside trainer.

Acceptance Criteria

-[ ] Add unit tests for functions in trainer

Add tests for extract_features.py

Description

Add a pytest script to test functions inside extract_features.py.

Acceptance Criteria

-[ ] Add unit tests for functions in extract_features.py

LPCC feature integration

LPC and LPCC feature integration.

• Add feature generation code
• Integrate with the classification pipeline
• Add tests