- Bug fixes
added use_beg_ms parameter in
soundpy.dsp.vad: improved VAD recognition of silences post speech.
added GPU option: provide instructions and Docker image for running SoundPy with GPU
added beg_end_clipped parameter to
soundpy.feats.plot_vadto visualize VAD by clipping the beginning and ending silences (if True) or VAD instances throughout the signal (if False).
soundpy.models.dataprep.GeneratorFeatExtractionclass for extracting and augmenting features during training (still experimental).
soundpy.models.builtin.envclassifier_extract_trainas an example of extracting and augmenting features during training (still experimental).
soundpy.dsp.clip_at_zeroto enable smoother concatenations of signals and enables removal of clicks at beginning and ending of signals.
soundpy.dsp.remove_dc_biasto enable smoother concatenations of signals
added mirror_sound option to
soundpy.dsp.apply_sample_lengthas a way to extend sound.
soundpy.dsp.ismonoto check if samples were mono or stereo.
soundpy.dsp.average_channelsto average sample amplitudes across channels, e.g. to identify where high energy begins / ends in the signal without disregarding additional channels (if stereo sound).
soundpy.dsp.add_channelsfor adding additional channels if needed (e.g. for applying a ‘hann’ or ‘hamming’ window to stereo sound)
- Other changes
name change: from pysoundtool to soundpy: simpler
updated dependencies to newest versions still compatible with Tensorflow 2.1.0
moved soundpy.dsp.get_vad_samples to
moved soundpy.dsp.get_vad_stft to
name change: allow
soundpy.feats.normalizeto be used as
removed pysoundtool_online and mybinder button as maintaining the online version was not easily done. Aim to reimplement at some point.
Initial public alpha release.