/**
* <JA>
* @brief 第1パス平行認識処理の初期化.
*
* MFCC計算のワークエリア確保を行う. また必要な場合は，スペクトル減算用の
* ワークエリア準備，ノイズスペクトルのロード，CMN用の初期ケプストラム
* 平均データのロードなども行われる.
*
* この関数は，システム起動後1回だけ呼ばれる.
* </JA>
* <EN>
* @brief Initializations for the on-the-fly 1st pass decoding.
*
* Work areas for all MFCC caculation instances are allocated.
* Additionaly,
* some initialization will be done such as allocating work area
* for spectral subtraction, loading noise spectrum from file,
* loading initial ceptral mean data for CMN from file, etc.
*
* This will be called only once, on system startup.
* </EN>
*
* @param recog [i/o] engine instance
*
* @callgraph
* @callergraph
*/
boolean
RealTimeInit(Recog *recog)
{
Value *para;
Jconf *jconf;
RealBeam *r;
MFCCCalc *mfcc;
jconf = recog->jconf;
r = &(recog->real);
/* 最大フレーム長を最大入力時間数から計算 */
/* set maximum allowed frame length */
r->maxframelen = MAXSPEECHLEN / recog->jconf->input.frameshift;
/* -ssload 指定時, SS用のノイズスペクトルをファイルから読み込む */
/* if "-ssload", load noise spectrum for spectral subtraction from file */
for(mfcc = recog->mfcclist; mfcc; mfcc = mfcc->next) {
if (mfcc->frontend.ssload_filename && mfcc->frontend.ssbuf == NULL) {
if ((mfcc->frontend.ssbuf = new_SS_load_from_file(mfcc->frontend.ssload_filename, &(mfcc->frontend.sslen))) == NULL) {
jlog("ERROR: failed to read \"%s\"\n", mfcc->frontend.ssload_filename);
return FALSE;
}
/* check ssbuf length */
if (mfcc->frontend.sslen != mfcc->wrk->bflen) {
jlog("ERROR: noise spectrum length not match\n");
return FALSE;
}
mfcc->wrk->ssbuf = mfcc->frontend.ssbuf;
mfcc->wrk->ssbuflen = mfcc->frontend.sslen;
mfcc->wrk->ss_alpha = mfcc->frontend.ss_alpha;
mfcc->wrk->ss_floor = mfcc->frontend.ss_floor;
}
}
for(mfcc = recog->mfcclist; mfcc; mfcc = mfcc->next) {
para = mfcc->para;
/* 対数エネルギー正規化のための初期値 */
/* set initial value for log energy normalization */
if (para->energy && para->enormal) energy_max_init(&(mfcc->ewrk));
/* デルタ計算のためのサイクルバッファを用意 */
/* initialize cycle buffers for delta and accel coef. computation */
if (para->delta) mfcc->db = WMP_deltabuf_new(para->baselen, para->delWin);
if (para->acc) mfcc->ab = WMP_deltabuf_new(para->baselen * 2, para->accWin);
/* デルタ計算のためのワークエリアを確保 */
/* allocate work area for the delta computation */
mfcc->tmpmfcc = (VECT *)mymalloc(sizeof(VECT) * para->vecbuflen);
/* MAP-CMN 用の初期ケプストラム平均を読み込んで初期化する */
/* Initialize the initial cepstral mean data from file for MAP-CMN */
if (para->cmn || para->cvn) mfcc->cmn.wrk = CMN_realtime_new(para, mfcc->cmn.map_weight);
/* -cmnload 指定時, CMN用のケプストラム平均の初期値をファイルから読み込む */
/* if "-cmnload", load initial cepstral mean data from file for CMN */
if (mfcc->cmn.load_filename) {
if (para->cmn) {
if ((mfcc->cmn.loaded = CMN_load_from_file(mfcc->cmn.wrk, mfcc->cmn.load_filename))== FALSE) {
jlog("WARNING: failed to read initial cepstral mean from \"%s\", do flat start\n", mfcc->cmn.load_filename);
}
} else {
jlog("WARNING: CMN not required on AM, file \"%s\" ignored\n", mfcc->cmn.load_filename);
}
}
}
/* 窓長をセット */
/* set window length */
r->windowlen = recog->jconf->input.framesize + 1;
/* 窓かけ用バッファを確保 */
/* set window buffer */
r->window = mymalloc(sizeof(SP16) * r->windowlen);
return TRUE;
}
/**
* <JA>
* 音声波形データから MFCC パラメータを抽出する．
*
* @param speech [in] 音声波形データ
* @param speechlen [in] @a speech の長さ（単位：サンプル数）
*
* @return 新たに割り付けられ抽出パラメータベクトルが格納されている
* パラメータ構造体へのポインタを返す．
* </JA>
* <EN>
* Extract MFCC parameters with sentence CMN from given waveform.
*
* @param speech [in] buffer of speech waveform
* @param speechlen [in] length of @a speech in samples
*
* @return pointer to newly allocated parameter structure data with extracted
* MFCC vector sequence.
* </EN>
*/
HTK_Param *new_wav2mfcc(SP16 speech[], int speechlen)
{
HTK_Param *param;
int framenum;
int i;
int len;
if (ssload_filename && ssbuf == NULL) {
/* load noise spectrum for spectral subtraction from file (once) */
if ((ssbuf = new_SS_load_from_file(ssload_filename, &sslen)) == NULL) {
j_error("Error: failed to read \"%s\"\n", ssload_filename);
}
}
if (sscalc) {
/* compute noise spectrum from head silence for each input */
len = sscalc_len * para.smp_freq / 1000;
if (len > speechlen) len = speechlen;
#ifdef SSDEBUG
printf("[%d]\n", len);
#endif
ssbuf = new_SS_calculate(speech, len, para, &sslen);
}
#ifdef SSDEBUG
{
int i;
for(i=0;i<sslen;i++) {
printf("%d: %f\n", i, ssbuf[i]);
}
}
#endif
/* calculate frame length from speech length, frame size and frame shift */
framenum = (int)((speechlen - para.framesize) / para.frameshift) + 1;
if (framenum < 1) {
j_printerr("input too short (%d samples), ignored\n", speechlen);
return NULL;
}
/* malloc new param */
param = new_param();
param->parvec = (VECT **)mymalloc(sizeof(VECT *) * framenum);
for(i=0;i<framenum;i++) {
param->parvec[i] = (VECT *)mymalloc(sizeof(VECT) * para.veclen);
}
/* make MFCC from speech data */
Wav2MFCC(speech, param->parvec, para, speechlen, ssbuf, sslen);
/* set miscellaneous parameters */
param->header.samplenum = framenum;
param->header.wshift = para.smp_period * para.frameshift;
param->header.sampsize = para.veclen * sizeof(VECT); /* not compressed */
param->header.samptype = F_MFCC;
if (para.delta) param->header.samptype |= F_DELTA;
if (para.acc) param->header.samptype |= F_ACCL;
if (para.energy) param->header.samptype |= F_ENERGY;
if (para.c0) param->header.samptype |= F_ZEROTH;
if (para.absesup) param->header.samptype |= F_ENERGY_SUP;
if (para.cmn) param->header.samptype |= F_CEPNORM;
param->veclen = para.veclen;
param->samplenum = framenum;
return param;
}
/**
* <JA>
* 音声波形データから MFCC パラメータを抽出する.
* エンジンインスタンス内の MFCC 計算インスタンスごとにパラメータ抽出が
* 行われ，それぞれの mfcc->param に格納される.
*
* @param speech [in] 音声波形データ
* @param speechlen [in] @a speech の長さ（単位：サンプル数）
* @param recog [in] エンジンインスタンス
*
* @return 成功時 TRUE, エラー時 FALSE を返す.
* </JA>
* <EN>
* Extract MFCC parameters with sentence CMN from given waveform.
* Parameters will be computed for each MFCC calculation instance
* in the engine instance, and stored in mfcc->param for each.
*
* @param speech [in] buffer of speech waveform
* @param speechlen [in] length of @a speech in samples
* @param recog [in] engine instance
*
* @return TRUE on success, FALSE on error.
* </EN>
*
* @callgraph
* @callergraph
*/
boolean
wav2mfcc(SP16 speech[], int speechlen, Recog *recog)
{
int framenum;
int len;
Value *para;
MFCCCalc *mfcc;
int veclen;
int t, i;
/* calculate frame length from speech length, frame size and frame shift */
framenum = (int)((speechlen - recog->jconf->input.framesize) / recog->jconf->input.frameshift) + 1;
if (framenum < 1) {
jlog("WARNING: input too short (%d samples), ignored\n", speechlen);
return FALSE;
}
for(mfcc=recog->mfcclist;mfcc;mfcc=mfcc->next) {
if (mfcc->frontend.ssload_filename) {
/* setup for spectral subtraction using file */
if (mfcc->frontend.ssbuf == NULL) {
/* load noise spectrum for spectral subtraction from file (once) */
if ((mfcc->frontend.ssbuf = new_SS_load_from_file(mfcc->frontend.ssload_filename, &(mfcc->frontend.sslen))) == NULL) {
jlog("ERROR: wav2mfcc: failed to read noise spectrum from file \"%s\"\n", mfcc->frontend.ssload_filename);
return FALSE;
}
}
}
if (mfcc->frontend.sscalc) {
/* compute noise spectrum from head silence for each input */
len = mfcc->frontend.sscalc_len * recog->jconf->input.sfreq / 1000;
if (len > speechlen) len = speechlen;
#ifdef SSDEBUG
jlog("DEBUG: [%d]\n", len);
#endif
mfcc->frontend.ssbuf = new_SS_calculate(speech, len, &(mfcc->frontend.sslen), mfcc->frontend.mfccwrk_ss, mfcc->para);
}
}
/* compute mfcc from speech file for each mfcc instances */
for(mfcc=recog->mfcclist;mfcc;mfcc=mfcc->next) {
para = mfcc->para;
veclen = para->veclen * mfcc->splice;
if (framenum - (mfcc->splice - 1) < 1) {
jlog("WARNING: input too short (%d samples), ignored\n", speechlen);
return FALSE;
}
/* malloc new param */
param_init_content(mfcc->param);
if (param_alloc(mfcc->param, framenum, veclen) == FALSE) {
jlog("ERROR: failed to allocate memory for converted parameter vectors\n");
return FALSE;
}
if (mfcc->frontend.ssload_filename || mfcc->frontend.sscalc) {
/* make link from mfccs to this buffer */
mfcc->wrk->ssbuf = mfcc->frontend.ssbuf;
mfcc->wrk->ssbuflen = mfcc->frontend.sslen;
mfcc->wrk->ss_alpha = mfcc->frontend.ss_alpha;
mfcc->wrk->ss_floor = mfcc->frontend.ss_floor;
}
/* make MFCC from speech data */
if (Wav2MFCC(speech, mfcc->param->parvec, para, speechlen, mfcc->wrk, mfcc->cmn.wrk) == FALSE) {
jlog("ERROR: failed to compute features from input speech\n");
if (mfcc->frontend.sscalc) {
free(mfcc->frontend.ssbuf);
mfcc->frontend.ssbuf = NULL;
}
return FALSE;
}
/* splicing */
if (mfcc->splice > 1) {
for (t = 0; t < framenum - (mfcc->splice - 1); t++) {
for (i = 1; i < mfcc->splice; i++) {
memcpy(&(mfcc->param->parvec[t][para->veclen * i]), &(mfcc->param->parvec[t + i][0]), sizeof(VECT) * para->veclen);
}
}
}
/* set miscellaneous parameters */
mfcc->param->header.samplenum = framenum - (mfcc->splice - 1);
mfcc->param->header.wshift = para->smp_period * para->frameshift;
mfcc->param->header.sampsize = veclen * sizeof(VECT); /* not compressed */
mfcc->param->header.samptype = para->basetype;
if (para->delta) mfcc->param->header.samptype |= F_DELTA;
if (para->acc) mfcc->param->header.samptype |= F_ACCL;
if (para->energy) mfcc->param->header.samptype |= F_ENERGY;
if (para->c0) mfcc->param->header.samptype |= F_ZEROTH;
if (para->absesup) mfcc->param->header.samptype |= F_ENERGY_SUP;
if (para->cmn) mfcc->param->header.samptype |= F_CEPNORM;
mfcc->param->veclen = veclen;
mfcc->param->samplenum = framenum - (mfcc->splice - 1);
if (mfcc->frontend.sscalc) {
free(mfcc->frontend.ssbuf);
mfcc->frontend.ssbuf = NULL;
}
}
return TRUE;
}
