Feature extraction

Contents

Feature extraction#

This submodule provides some feature extraction utilities that can be used as key_transform functions in MLX data pipelines. Even though a C++ implementation would allow for completely circumventing the GIL and better utilization of multiple threads, we find that an efficient numpy implementation can often be fast enough while providing signficiantly more flexibility.

Audio Features#

WindowType(value[, names, module, qualname, ...])

Enum to choose the window function.

FrequencyScale(value[, names, module, ...])

Enum to choose the frequency scaling for the filter banks.

mfsc(n_filterbank, sampling_freq[, ...])

Returns a function that computes spectrogram features from audio in particular mel-frequency spectral coefficients (MFSCs).