fe_endpoint.h

这是一个语音特征提取的程序源码

///////////////////////////////////////////////////////////////////////////////
// This is a part of the Feature program.
// Version: 1.0
// Date: February 22, 2003
// Programmer: Oh-Wook Kwon
// Copyright(c) 2003 Oh-Wook Kwon. All rights reserved. [e-mail]
///////////////////////////////////////////////////////////////////////////////

#ifndef _FE_ENDPOINT_H_
#define _FE_ENDPOINT_H_


#include "FE_common.h"


/*-----------------------------------*/
/* Definition for endpoint detectors */
/*-----------------------------------*/
#define EPD_MAX_RECORD_TIME      10           /* 10 seconds */
#define EPD_WAV_BUF_SIZE         (EPD_MAX_RECORD_TIME*16000)   /* 10 seconds at 16 kHz mono sampling frequency */
#define EPD_FRAME_SHIFT_IN_MS    10           /* frame shift in millisecond (10 ms) */
#define EPD_MAX_WIN_SIZE         512          /* maximum window size */


/*-----------------------------------*/
/* Definition for endpoint detectors */
/*-----------------------------------*/
#define EPD_MAX_PAUSE_IN_MS      2000         /* maximum duration of long-pause in millisecond (<=2000 ms) */


/*-----------------*/
/* Type definition */
/*-----------------*/
typedef enum {
	EPD_STATUS_WAITING,
		EPD_STATUS_SPEECH,
		EPD_STATUS_PAUSE,
		EPD_STATUS_ENDPOINT
} EpdLocalStatus;


typedef enum {
	EPD_FK_SILENCE=0,
		EPD_FK_SPEECH
} EpdFrameKind;


/*----------------*/
/* Data structure */
/*----------------*/

typedef struct {
	int   m_sampleRate;
	int   m_shiftSize; /* Shift size in samples (10 ms) */
	int   m_winSize; /* Window size in samples, same as feature extraction. */
	int   m_threshFrameN;
	int   m_startFrameN;
	int   m_endFrameN;
	int   m_startSilenceFrameN;
	int   m_endSilenceFrameN;
	int   m_longPauseFrameN; /* to detect utterance end-point */
} EpdParm;


#ifdef _DEBUG
#define EPD_SPEECH_BUF_SIZE         EPD_WAV_BUF_SIZE
#define EPD_FRAME_BUF_SIZE          (EPD_MAX_RECORD_TIME*100) /* assuming 10 ms frame shift */
#else
#define EPD_SPEECH_BUF_SIZE         ((EPD_MAX_PAUSE_IN_MS/10+1)*EPD_MAX_WIN_SIZE)  /* to save memory */
#define EPD_FRAME_BUF_SIZE          (EPD_MAX_PAUSE_IN_MS/10+1)  /* to save memory */
#endif

#define EPD_BUF_SIZE                7


class Epd {
public:
	/* fixed part */
	int   m_isActive;
	int   m_isAudio;
	EpdParm m_config;

	/* adaptive part */
	long m_uttBeginX; /* start sample point of speech */
	long m_uttEndX; /* end sample point of speech */
	long m_localFrameX; /* local time frame index */

	/* adaptive noise estimation */
	float m_lambdaLTE;
	float m_noiseEn;
	float m_lambdaLTEhigherE;
	int m_nbSpeechFrame;
	int m_nbFrameEpd;
	EpdFrameKind m_flagVAD;

	/* ZCR tracking */
	float m_noiseLevel;
	float m_lambdaZcr;
	float m_meanZcr;

	/* SNR tracking */
	float m_signalEn;
	float m_lambdaSignalE;
	long m_absTimeX;
	float m_lastSnr;

	/* working variables */
	EpdLocalStatus m_localStatus; /* The status of EPD must be hidden to other modules */
	EpdFrameKind m_isSpeechA[EPD_FRAME_BUF_SIZE];
	float m_zcrA[EPD_FRAME_BUF_SIZE]; /* to remove breath noise */
	long m_sampleEndX; /* end sample point to input audio */
	short m_epdSpeech[EPD_SPEECH_BUF_SIZE];
	int m_speechSegN; /* number of detected speech segments */

	/*--------------------*/
	/* Member functions */
	/*--------------------*/
	
	Epd();
	virtual ~Epd();
	int Init(int samplingRate, int isAudio, int isActive, int isDenoised);
	int InitNewUtterance();
	FeReturnCode OneFrame(float *in, float *out, int frameX, int winSize, FeReturnCode inStatus, EpdFrameKind *frameKind);
	int GetOutput(short *sampleA, int maxSampleN);
	void Close();
	void SetMaxPause(int msec);
#ifdef _DEBUG
	int SaveOutput(const char *fname, int offsetX);
#endif

private:
	int ParmInit(EpdParm *epdParm, int samplingRate, int isDenoised);
	int PutSample(float *sampleA, int sampleN);
	int FindBeginPoint(int startX);
	int FindEndPoint(int startX, int reqSilN);
	int SaveAudio(const char *fname, int begX, int endX);
	EpdFrameKind OneFrame(const float *s);

};


#endif