hparm.h

该压缩包为最新版htk的源代码,htk是现在比较流行的语音处理软件,请有兴趣的朋友下载使用

/* ----------------------------------------------------------- */
/*                                                             */
/*                          ___                                */
/*                       |_| | |_/   SPEECH                    */
/*                       | | | | \   RECOGNITION               */
/*                       =========   SOFTWARE                  */ 
/*                                                             */
/*                                                             */
/* ----------------------------------------------------------- */
/* developed at:                                               */
/*                                                             */
/*      Speech Vision and Robotics group                       */
/*      Cambridge University Engineering Department            */
/*      http://svr-www.eng.cam.ac.uk/                          */
/*                                                             */
/*      Entropic Cambridge Research Laboratory                 */
/*      (now part of Microsoft)                                */
/*                                                             */
/* ----------------------------------------------------------- */
/*         Copyright: Microsoft Corporation                    */
/*          1995-2000 Redmond, Washington USA                  */
/*                    http://www.microsoft.com                 */
/*                                                             */
/*              2001  Cambridge University                     */
/*                    Engineering Department                   */
/*                                                             */
/*   Use of this software is governed by a License Agreement   */
/*    ** See the file License for the Conditions of Use  **    */
/*    **     This banner notice must not be removed      **    */
/*                                                             */
/* ----------------------------------------------------------- */
/*         File: HParm.h:   Speech Parameter Input/Output      */
/* ----------------------------------------------------------- */

/* !HVER!HParm:   3.3 [CUED 28/04/05] */

#ifndef _HPARM_H_
#define _HPARM_H_

#ifdef __cplusplus
extern "C" {
#endif

enum _BaseParmKind{
      WAVEFORM,            /* Raw speech waveform (handled by HWave) */
      LPC,LPREFC,LPCEPSTRA,LPDELCEP,   /* LP-based Coefficients */
      IREFC,                           /* Ref Coef in 16 bit form */
      MFCC,                            /* Mel-Freq Cepstra */
      FBANK,                           /* Log Filter Bank */
      MELSPEC,                         /* Mel-Freq Spectrum (Linear) */
      USER,                            /* Arbitrary user specified data */
      DISCRETE,                        /* Discrete VQ symbols (shorts) */
      PLP,                             /* Standard PLP coefficients */
      ANON};
      
typedef short ParmKind;          /* BaseParmKind + Qualifiers */
                                 
#define HASENERGY  0100       /* _E log energy included */
#define HASNULLE   0200       /* _N absolute energy suppressed */
#define HASDELTA   0400       /* _D delta coef appended */
#define HASACCS   01000       /* _A acceleration coefs appended */
#define HASCOMPX  02000       /* _C is compressed */
#define HASZEROM  04000       /* _Z zero meaned */
#define HASCRCC  010000       /* _K has CRC check */
#define HASZEROC 020000       /* _0 0'th Cepstra included */
#define HASVQ    040000       /* _V has VQ index attached */
#define HASTHIRD 0100000       /* _T has Delta-Delta-Delta index attached */

#define BASEMASK  077         /* Mask to remove qualifiers */

/*
   An observation contains one or more stream values each of which 
   is either a vector of continuous values and/or a single
   discrete symbol.  The discrete vq symbol is included if the
   target kind is DISCRETE or the continuous parameter has the
   HASVQ qualifier. Observations are input via buffers or tables.  A
   buffer is a FIFO structure of potentially infinite length and it is
   always sourced via HAudio.  A table is a random access array of
   observations and it is sourced from a file possibly via HWave.
   Buffers are input only, a table can be input and output.
   Too allow discrete systems to be used directly from continuous
   data the observation also holds a separate parm kind for the
   parm buffer and routines which supply observations use this to
   determine stream widths when the observation kind is DISCRETE.
*/

typedef enum {
   FALSE_dup=FALSE, /*  0 */
   TRUE_dup=TRUE,   /*  1 */
   TRI_UNDEF=-1     /* -1 */
}
TriState;

typedef struct {
   Boolean eSep;         /* Energy is in separate stream */
   short swidth[SMAX];   /* [0]=num streams,[i]=width of stream i */
   ParmKind bk;          /* parm kind of the parm buffer */
   ParmKind pk;          /* parm kind of this obs (bk or DISCRETE) */
   short vq[SMAX];       /* array[1..swidth[0]] of VQ index */
   Vector fv[SMAX];      /* array[1..swidth[0]] of Vector */
} Observation;

/*
   A ParmBuf holds either a static table of parameter frames
   loaded from a file or a potentially infinite sequence
   of frames from an audio source. The key information relating 
   to the speech data in a buffer or table can be obtained via 
   a BufferInfo Record.  A static table behaves like a stopped
   buffer.
*/

typedef enum { 
   PB_INIT,     /* Buffer is initialised and empty */
   PB_WAITING,  /* Buffer is waiting for speech */
   PB_STOPPING, /* Buffer is waiting for silence */
   PB_FILLING,  /* Buffer is filling */
   PB_STOPPED,  /* Buffer has stopped but not yet empty */
   PB_CLEARED   /* Buffer has been emptied */
} PBStatus;

typedef struct _ParmBuf  *ParmBuf;

typedef struct {
   ParmKind srcPK;            /* Source ParmKind */ 
   FileFormat srcFF;          /* Source File format */ 
   HTime srcSampRate;         /* Source Sample Rate */ 
   int srcVecSize;            /* Size of source vector */
   long nSamples;             /* Number of source samples */
   int frSize;                /* Number of source samples in each frame */
   int frRate;                /* Number of source samples forward each frame */
   int nObs;                  /* Number of table observations */
   ParmKind tgtPK;            /* Target ParmKind */ 
   FileFormat tgtFF;          /* Target File format */ 
   HTime tgtSampRate;         /* Target Sample Rate */ 
   int tgtVecSize;            /* Size of target vector */
   AudioIn a;                 /* the audio source - if any */
   Wave w;                    /* the wave input - if any */
   Ptr i;                     /* the other input - if any */
   Boolean useSilDet;         /* Use Silence Detector */
   int audSignal;             /* Signal Number for Audio Control */
   char *vqTabFN;             /* Name of VQ Table Defn File */
   Boolean saveCompressed;    /* Save in compressed format */
   Boolean saveWithCRC;       /* Save with CRC check added */
   Boolean spDetParmsSet;     /* Parameters set for sp/sil detector */
   float spDetSil;            /* Silence level for channel */
   float chPeak;              /* Peak-to-peak input level for channel */
   float spDetSp;             /* Speech level for channel */
   float spDetSNR;            /* Speech/noise ratio for channel */
   float spDetThresh;         /* Silence/speech level threshold */
   float curVol;              /* Volume level of last frame (0.0-100.0dB) */
   int spDetSt;               /* Frame number of first frame of buffer */
   int spDetEn;               /* Frame number of last frame of buffer */
   char *matTranFN;           /* Matrix transformation name */
}BufferInfo;

/*
   External source definition structure
*/

typedef struct hparmsrcdef *HParmSrcDef;

/* -------------------- Initialisation ------------------ */

ReturnStatus InitParm(void);
/*
   Initialise the module
*/

/* -------------------- Channel functions ------------------ */

ReturnStatus SetChannel(char *chanName);
/* 
   Set the current channel to use config parameters from chanName.
*/

void ResetChannelSession(char *chanName);
/* 
   Reset the session for the specified channel (NULL indicates default)
*/

/* 
   The next two functions have been kept to allow for backwards 
   compatibility.
*/
void SetNewConfig(char * libmod);
void ResetCurCepMean(void);

/* ---------------- Buffer Input Routines ------------------ */

ParmBuf OpenBuffer(MemHeap *x, char *fn, int maxObs, FileFormat ff, 
		   TriState enSpeechDet, TriState silMeasure);
/*
   Open and return a ParmBuf object connected to the current channel.
   If maxObs==0 blocks and reads whole of file/audio into memory and
   returns with status==PB_STOPPED ready for table access.  All 
   parameters, associated with the loading and conversion of the
   source are defined using configuration parameters.
   If maxObs!=0 buffer may be read as a stream.  In this case reading
   should be via ReadAsBuffer calls which should continue until either
   ReadAsBuffer returns FALSE or buffer status >= PB_STOPPED.  Note 
   that for some types of input (eg pipes) end of data can only be 
   determined by a failed attempt to read the final frame.
   If the speech detector is enabled (either by configuration or
   by explicit parameter in call) then silence measurement can be
   forced/prevented by setting silMeasure to TRUE/FALSE (if UNDEF
   will perform measurement if it is needed by config).