diffsptk version 1.0.0

What's Changed

• Purge freqt2 from smcep by @takenori-y in #26
• Support zero-phase non-cascade MLSA filter by @takenori-y in #28
• Add yingram by @takenori-y in #29
• Add mlsacheck by @takenori-y in #30
• Expose PQMF parameters by @takenori-y in #33
• Support sawtooth excitation generation by @takenori-y in #37
• Support MLSA filter in the frequency domain by @takenori-y in #36
• Improve linear_intpl by @yoyolicoris in #38
• Add gmm by @takenori-y in #39
• Support learnable PQMF by @takenori-y in #41
• Add ap by @takenori-y in #43
• Add poledf by @takenori-y in #44
• Add rlevdur by @takenori-y in #45
• Add magic_intpl by @takenori-y in #47

Major Breaking Changes

• cascade parameter in MLSA has been changed to mode parameter
• Minimal PyTorch version has been changed to 1.11.0

diffsptk version 0.6.0

What's Changed

• Support sinusoidal excitation by @takenori-y in #15
• Get alpha from sample rate by @takenori-y in #16
• Add root_pol by @takenori-y in #17
• Add vq by @takenori-y in #19
• Add msvq and imsvq by @takenori-y in #20
• Add unframe by @takenori-y in #21
• Add istft by @takenori-y in #22
• Add lbg by @takenori-y in #23
• Fix slow import by @takenori-y in #24
• Module sopr was stripped

diffsptk version 0.5.0

Fifth release

Additional implemented modules are:

• cqt
• entropy
• excite
• imglsadf
• mglsadf
• train

The recursive computation used in c2mpir, mpir2c, mgc2mgc, and mgc2sp has been changed to the FFT-based one for parallel processing.

diffsptk version 0.4.0

Fourth release

Additional implemented modules are:

• delay
• lar2par
• mgcep (updated from mcep)
• par2lar
• pca
• pitch
• smcep
• sopr

Note pca requires PyTorch 1.10+. Currently, only 16 kHz is supported in pitch. MLSA filter will be implemented in the next release.

diffsptk version 0.3.0

Third release

Some minor bugs were fixed by introducing a coverage test.

Additional implemented modules are:

• df2
• dfs
• grpdelay
• lpccheck
• mcpf
• mpir2c
• phase

The default floor values of fbank and mfcc were changed.

diffsptk version 0.2.0

Second release

Additional implemented modules are:

• c2ndps
• delta
• dequantize
• gnorm
• ignorm
• lpc2par
• mgc2mgc
• mgc2sp
• mlpg
• ndps2c
• norm0
• par2lpc
• quantize
• rmse
• snr
• zcross

Currently, mlpg assumes a global diagonal unit variance. Note that zcross cannot compute gradient.

diffsptk version 0.1.0

Initial release

Implemented modules are:

• acorr
• b2mc
• c2mpir
• cdist
• dct
• decimate
• fbank
• fftcep
• frame
• freqt
• idct
• interpolate
• ipqmf
• iulaw
• levdur
• linear_intpl
• lpc
• mc2b
• mfcc
• mcep
• pqmf
• spec
• stft
• ulaw
• window
• zerodf

Currently, levdur and mcep solve their linear system by naive matrix inversion.