SPTK Version 4.0

Major Changes from SPTK3

• Signal processing tools were written in C++ instead of C
• Input and output types were changed to double from float
• Drawing commands were implemented in Python
• Following commands were obsoleted
  • acep, agcep, and amcep
  • bell
  • c2sp
  • cat2 and echo2
  • da
  • ds, us, us16, and uscd
  • fig
  • gc2gc
  • gcep, mcep, and uels
  • glsadf, lmadf, and mlsadf
  • ivq and vq
  • lsp2sp
  • mgc2mgclsp
  • psgr and xgr
  • raw2wav, wav2raw, wavjoin, and wavsplit
• Following commands were separated
  • dtw (dtw and dtw_merge)
  • mglsadf (mglsadf and imglsadf)
  • train (train and mseq)
  • ulaw (ulaw and iulaw)
  • vstat (vstat and median)
• Following commands were renamed
  • c2ir to c2mpir
  • mgclsp2sp to mglsp2sp

New Features

• Conversion from/to log area ratio (lar2par and par2lar)
• Entropy calculation (entropy)
• Huffman coding (huffman, huffman_encode, and huffman_decode)
• Mel-cepstrum postfilter (mcpf)
• Mel-filter-bank extraction (fbank)
• Nonrecursive MLPG (mlpg -R 1)
• Pitch extraction by DIO used in WORLD (pitch -a 3)
• Scalar quantization (quantize and dequantize)
• Stability check of LPC coefficients (lpccheck)
• Subband decomposition (pqmf and ipqmf)
• Windows build support (only static library)

Acknowledgments

• Keiichi Tokuda @k1tokuda
• Keiichiro Oura @uratec
• Takenori Yoshimura @takenori-y
• Takato Fujimoto @taka19