gossp

GOSSP

GOSSP is a Speech Signal Processing library for the Go language, which includes time-frequency analysis, fundamental frequency estimation, spectral envelope estimation and waveform generation filters, etc.

Packages

• cwt - Continuous Wavelet Tranform (CWT) and inverse CWT (in develop).
• dct - Discrete Cosine Transform (DCT) and Inverse DCT.
• dtw - Dynamic Time Warping (DTW)
• excite - Excitation generation from fundamental frequency.
• f0 - Fundamental frequency (f0) estimatnion.
• io - Input/Output (in develop).
• mgcep - Mel-generalized cepstrum analysis for spectral envelope estimation.
• special - Special functions analogy to scipy in python.
• stft - Short-Time Fourier Transform (STFT) and Inverse STFT.
• vocoder - Speech waveform generation filters.
• window - Window functions.
• z - Z-transform to analyze digital filters.

Install

 go get github.com/r9y9/gossp

Install SPTK

To use SPTK with GOSSP, you need to install the modified version of SPTK as follows:

 git clone https://github.com/r9y9/SPTK.git && cd SPTK
 ./waf configure && ./waf
 sudo ./waf install

Getting Started

Short-Time Fourier Transform

package main

import (
	"flag"
	"fmt"
	"github.com/r9y9/gossp"
	"github.com/r9y9/gossp/io"
	"github.com/r9y9/gossp/stft"
	"github.com/r9y9/gossp/window"
	"log"
	"math"
)

func main() {
	filename := flag.String("i", "input.wav", "Input filename")
	flag.Parse()

	w, werr := io.ReadWav(*filename)
	if werr != nil {
		log.Fatal(werr)
	}
	data := w.GetMonoData()

	s := &stft.STFT{
		FrameShift: int(float64(w.SampleRate) / 100.0), // 0.01 sec,
		FrameLen:   2048,
		Window:     window.CreateHanning(2048),
	}

	spectrogram, _ := gossp.SplitSpectrogram(s.STFT(data))
	PrintMatrixAsGnuplotFormat(spectrogram)
}

func PrintMatrixAsGnuplotFormat(matrix [][]float64) {
	fmt.Println("#", len(matrix[0]), len(matrix)/2)
	for i, vec := range matrix {
		for j, val := range vec[:1024] {
			fmt.Println(i, j, math.Log(val))
		}
		fmt.Println("")
	}
}

Waveform Reconstruction using Inverse Short-Time Fourier Transform

package main

import (
	"flag"
	"fmt"
	"github.com/r9y9/gossp/io"
	"github.com/r9y9/gossp/stft"
	"github.com/r9y9/gossp/window"
	"log"
)

func main() {
	filename := flag.String("i", "input.wav", "Input filename")
	flag.Parse()

	w, werr := io.ReadWav(*filename)
	if werr != nil {
		log.Fatal(werr)
	}
	data := w.GetMonoData()

	s := &stft.STFT{
		FrameShift: int(float64(w.SampleRate) / 100.0), // 0.01 sec,
		FrameLen:   2048,
		Window:     window.CreateHanning(2048),
	}

	spectrogram := s.STFT(data)

	// do something on spectrogram

	reconstructed := s.ISTFT(spectrogram)

	for i := range data {
		// expect to be same
		fmt.Println(data[i], reconstructed[i])
	}
}

Fun with speech signal processing!

Third Party Libraries

• GO-DSP
• MCEP Alpha Calculator
• SPTK

LICENSE

MIT