Numpy/SciPy-based implementation of cross-correlation for sound-source localization
To calibrate,
- record sound files (wav format) with sound sources at different angles from your two-microphone system. Note the angle to the sound sources in the file names.
- Update training_config.yaml; in particular update the regex for extracting the angle to the sound source from the sound files' names.
- Run train.py to generate a calibration file.
That calibration file can then be used with crosscorrelizer.py to estimate the angle to a sound source from new wav files recorded with the same two-microphone system.