A small package to generate features from acoustic
Project description
B2AI Prep
A simple Python package to prepare acoustic data for the Bridge2AI voice project.
Caution: this package is still under development and may change rapidly over the next few weeks.
Installation
Requires a Python >= 3.10 environment
pip install b2aiprep
Usage
Two commands are available through the CLI:
b2aiprep-cli --help
- Convert an audio file to features:
The simplest form takes an audio file, a subject id, and a task name.
b2aiprep-cli convert test_audio.wav s1 mpt
It will save a pytorch .pt
file with a dictionary of features. This can be
loaded by torch.load()
. The file is named following a simple convention:
sub-<subject_id>_task-<task_name>_md5-<checksum>_features.pt
- Batch process audio files
This requires a CSV file, where each line is of the form: path/to/audio.wav,subject_id,task_name
b2aiprep-cli batchconvert filelist.csv --plugin cf n_procs=2 --outdir out
The above command uses pydra under the hood to parallel process the audio files.
All outputs are currently stored in a single directory specified by the --outdir
flag.
- Verify if two audio files are from the same speaker
b2aiprep-cli test_audio1.wav test_audio2.wav --model 'speechbrain/spkrec-ecapa-voxceleb'
This will use the speechbrain speaker recognition model to verify that the two audio files are from the same speaker.
There is a notebook in the docs directory that can be used to interact with the library programmatically.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.