/***************************************************************************
preprocess.h - Preprocessing of a wave file for
speech recognition
-------------------
begin : Sat Feb 12 2000
copyright : (C) 2000 by Daniel Kiecza
email : daniel@kiecza.de
***************************************************************************/
/***************************************************************************
* *
* This program is free software; you can redistribute it and/or modify *
* it under the terms of the GNU General Public License as published by *
* the Free Software Foundation; either version 2 of the License, or *
* (at your option) any later version. *
* *
***************************************************************************/
#ifndef PREPROCESS_H
#define PREPROCESS_H
/********************************************************************************
* fft_size Size of short-time window
* fft_size_char Size of short-time window (in number of char values)
*
* fft_size is size in number of 16-bit samples,
* a 16-bit sample is a two-byte value, thus:
* fft_size == 2 * fft_size_char
*
* as the power spectrum is symmetric, we have:
* power_spec_size = 0.5 * fft_size
********************************************************************************/
#define FFT_SIZE 256
#define POWER_SPEC_SIZE 128
#define FFT_SIZE_CHAR 512
#define HAMMING_SIZE FFT_SIZE
#define VECSIZE 16
#define FEAT_VEC_SIZE VECSIZE
/********************************************************************************
* Offset = 10ms (160 16bit-values = 320 uchars)
*
* the distance that lies between two audio frames
* generally, it is offset < fft_size, and thus
* subsequent frames do overlap!
********************************************************************************/
#define OFFSET 320
/*****
used for reduction of short-time spectrum to mel scale coefficients
*****/
int filter_banks[17];
/*****
contains the power spectrum
*****/
float power_spec[POWER_SPEC_SIZE];
/********************************************************************************
* Hamming window width = 16ms ! (256 Frames)
* (hamming_size == fft_size)
*
* a window function that a frame of audio values is multiplied with.
* This is done to smoothen the beginning and end of a frame,
* i.e. to reduce discontinuities at the two ends and as a result
* to reduce the number of artefacts in the power spectrum
********************************************************************************/
float hamming_window[HAMMING_SIZE];
/*****
The characteristics of the recording channel
This is substracted from each feature vector to reduce
channel effects
*****/
int do_mean_sub;
float channel_mean[FEAT_VEC_SIZE];
int initPreprocess();
int preprocessFrame(float *frame, float *result);
void endPreprocess();
#endif
