SPTK

The Speech Signal Processing Toolkit (SPTK) is a software for speech signal processing tools.

• Older version: SPTK3
• PyTorch version: diffsptk

What is SPTK?

• SPTK consists of over 100 commands for speech signal processing.
• The data format used in SPTK is raw header-less, i.e., there is no specific structure. Thanks to the data format, we can check file contents immediately on CUI.

  dmp +s data.raw

• The data used in the commands is passed through standard input/output. We can chain multiple processes using pipes.

  x2x +sd < data.raw | clip | x2x +da | less

• The data type is basically little-endian double 8 bytes.
• The commands do not require interactive user inputs. Parameters are set via command line options beforehand.

  impulse -l 4 | sopr -m 10 | x2x +da

Documentation

• See this page for a reference manual.
• Our paper is available on OpenReview.

Requirements

• GCC 4.8.5+ / Clang 3.5.0+ / Visual Studio 2015+
• CMake 3.1+

Installation

Linux / macOS

git clone https://github.com/sp-nitech/SPTK.git
cd SPTK
make

Windows

cd /path/to/SPTK  # Please change here to your appropriate path.
mkdir build
cd build
cmake .. -DCMAKE_INSTALL_PREFIX=..  # Please change install directory.
MSBuild /p:Configuration=Release INSTALL.vcxproj

Demonstration

• Twitter
• Analysis-synthesis via mel-cepstrum
• Parametric coding via line spectral pairs

Examples

SPTK provides some examples. Go to an example directory and execute run.sh, e.g.,

cd egs/analysis_synthesis/mgc
./run.sh

The below is a simple example that decreases the volume of input audio in input.wav. You may need to install sox command on your system.

sox -t wav input.wav -c 1 -t s16 -r 16000 - |
    x2x +sd | sopr -m 0.5 | x2x +ds -r |
    sox -c 1 -t s16 -r 16000 - -t wav output.wav

If you would like to draw figures, please prepare a python environment.

cd tools; make venv PYTHON_VERSION=3.8; cd ..
. ./tools/venv/bin/activate
impulse -l 32 | gseries impulse.png
deactivate

Changes from SPTK3

• Input and output types are changed to double from float
• Signal processing classes are written in C++ instead of C
• Drawing commands are implemented in Python
• No memory leaks
• Thread-safe
• New features:
  • Aperiodicity extraction (ap)
  • Conversion from/to log area ratio (lar2par and par2lar)
  • Dynamic range compression (drc)
  • Entropy calculation (entropy)
  • Huffman coding (huffman, huffman_encode, and huffman_decode)
  • Magic number interpolation (magic_intpl)
  • Median filter (medfilt)
  • Mel-cepstrum postfilter (mcpf)
  • Mel-filter-bank extraction (fbank)
  • Nonrecursive MLPG (mlpg -R 1)
  • Pitch adaptive spectrum estimation (pitch_spec)
  • Pitch extraction by DIO used in WORLD (pitch -a 3)
  • Pole-zero plot (gpolezero)
  • Scalar quantization (quantize and dequantize)
  • Spectrogram plot (gspecgram)
  • Stability check of LPC coefficients (lpccheck)
  • Subband decomposition (pqmf and ipqmf)
  • WORLD synthesis (world_synth)
  • Windows build support (only static library)
• Obsoleted commands:
  • acep, agcep, and amcep -> amgcep
  • bell
  • c2sp -> mgc2sp
  • cat2 and echo2
  • da
  • ds, us, us16, and uscd -> sox
  • fig
  • gc2gc -> mgc2mgc
  • gcep, mcep, and uels -> mgcep
  • glsadf, lmadf, and mlsadf -> mglsadf
  • ivq and vq -> imsvq and msvq
  • lsp2sp -> mglsp2sp
  • mgc2mgclsp and mgclsp2mgc
  • psgr and xgr
  • raw2wav, wav2raw, wavjoin, and wavsplit -> sox
• Separated commands:
  • c2ir -> c2mpir and mpir2c
  • dtw -> dtw and dtw_merge
  • mglsadf -> mglsadf and imglsadf
  • train -> train and mseq
  • ulaw -> ulaw and iulaw
  • vstat -> vstat and median
• Renamed commands:
  • mgclsp2sp -> mglsp2sp

Who we are

• Keiichi Tokuda - Produce and Design - Nagoya Institute of Technology
• Keiichiro Oura - Nagoya Institute of Technology
• Takenori Yoshimura - Main Maintainer - Nagoya Institute of Technology
• Takato Fujimoto - Nagoya Institute of Technology

Contributors to former versions of SPTK

• Akira Tamamori
• Cassia Valentini
• Chiyomi Miyajima
• Fernando Gil Resende Junior
• Gou Hirabayashi
• Heiga Zen
• Junichi Yamagishi
• Kazuhito Koishida
• Keiichi Tokuda
• Keiichiro Oura
• Kenji Chiba
• Masatsune Tamura
• Naohiro Isshiki
• Noboru Miyazaki
• Satoshi Imai
• Shinji Sako
• Tadashi Kitamura
• Takao Kobayashi
• Takashi Masuko
• Takashi Nose
• Takato Fujimoto
• Takayoshi Yoshimura
• Takenori Yoshimura
• Toru Takahashi
• Toshiaki Fukada
• Toshihiko Kato
• Toshio Kanno
• Yoshihiko Nankaku

License

This software is released under the Apache License 2.0.

Reference

@InProceedings{sp-nitech2023sptk,
  author = {Takenori Yoshimura and Takato Fujimoto and Keiichiro Oura and Keiichi Tokuda},
  title = {{SPTK4}: An open-source software toolkit for speech signal processing},
  booktitle = {12th ISCA Workshop on Speech Synthesis (SSW 12)},
  year = {2023},
}