This pr brings Whisper's feature extractor to Relax VM. The feature extractor extracts mel-filter bank features from raw speech data. The deviation between the current implementation and huggingface's WhisperFeatureExtractor._np_extract_fbank_features is less than 1e-4.
This pr brings Whisper's feature extractor to Relax VM. The feature extractor extracts mel-filter bank features from raw speech data. The deviation between the current implementation and huggingface's
WhisperFeatureExtractor._np_extract_fbank_featuresis less than 1e-4.