kapre
kapre copied to clipboard
TODO - 2020 version
- [x] Add inverse STFT
- [x] Add a
composed
layer that concatenates magnitude and phase of input complex number (which would be STFT)
- [x] mfcc?
- [ ] time-stretch?
- [ ] frequency shift?
- [x] fix test for phase
- [x] energy
- [ ] loudness
- [ ] HPSS
- [x] tf.signal.frame
- [x] mu law
- [ ] pcen
- [ ] freq and time mask (spec aug)
- [x] augmentation - swap channels
- [ ] augmentation - one for my ldb
- [ ] compute loudness like https://github.com/magenta/ddsp/blob/0c8c151cbddefb32d04c1b017f060ac44997d036/ddsp/spectral_ops.py#L173, which involves a-weighting (so we need backend for it)
- [x] frequency-aware conv layer? (@kkoutini any chance you'd be interested in making a PR?)
- [ ]
InstaneousFrequency(pad_end=True)
- [ ] pre-emphasis, de-emphasis
- [ ] LPC
- [ ] ERB
- [ ] LFCC
- [x] change numpy requirements to 1.18.5 to be compatible with tensorflow.
- [x] enable model name for composed layers
typos .. retunrs
- [ ] random gain (control the multiplication factor)
- [ ] random volume (control the output.. amplitude? energy?)
-
[ ] more examples
-
[ ] pre-structured models?