- Standalone example on microcontroller. XOR?
- Setup documentation build (Sphinx?)
- Consolidate error handling. Use EmlError, EML_PRE/POSTCONDITION, EML_CHECK_ERROR everywhere
- Implement some iterators like EML_APPLY(vector, 0, vector->length, func) and EML_MAP_INTO(in, out, 0, out->length, func);
- Add tests for melspectrogram
- Unhardcode FFT length
- Add some pipeline/processor. Short audio buffers in, feature extraction, normalization, classifier, predictions out.
- Add an example. Wakeword detection? Keyword classification?
- Implement low-level features. RMS, zero-crossings, spectral centroid, spectral flatness
- Add support for CNNs. 1D+2D convolutions, pooling (Keras).
- Add support for RNNs. SimpleRNN, LSTM, GRU
- Support quantized models (8 bits)
- Supports strides and dilation in convolution layers
- Use floats by default
- Support sklearn GradientBoostingClassifier
- Support regression trees
- Support weighted voting
- Implement Isolation Forests (requires path/depths)