int refineBeat(int isBeat)
{
BOOL accepted=FALSE;
BOOL predicted=FALSE;
BOOL resyncin=FALSE;
BOOL resyncout=FALSE;
if (isBeat) // Show the beat received from AVS
SliderStep(IDC_IN, &inSlide);
DWORD TCNow = GetTickCount();
if (songChanged(TCNow))
{
bestConfidence=(int)((float)bestConfidence*0.5);
sticked=0;
stickyConfidenceCount=0;
if (cfg_smartbeatresetnewsong)
ResetAdapt();
}
// Try to predict if this frame should be a beat
if (Bpm && TCNow > predictionLastTC + (60000 / Bpm))
predicted = TRUE;
if (isBeat) // If it is a real beat, do discrimination/guessing and computations, then see if it is accepted
accepted = TCHistStep(TCHist, Avg, halfDiscriminated, &hdPos, &lastTC, TCNow, BEAT_REAL);
// Calculate current Bpm
CalcBPM();
// If prediction Bpm has not yet been set
// or if prediction bpm is too high or too low
// or if 3/4 of our history buffer contains beats within the range of typical drift
// the accept the calculated Bpm as the new prediction Bpm
// This allows keeping the beat going on when the music fades out, and readapt to the new beat as soon as
// the music fades in again
if ((accepted || predicted) && !sticked && (!predictionBpm || predictionBpm > MAX_BPM || predictionBpm < MIN_BPM))
{
if (Confidence >= bestConfidence)
{
/* betterConfidenceCount++;
if (!predictionBpm || betterConfidenceCount == BETTER_CONF_ADOPT)
{*/
forceNewBeat=1;
/* betterConfidenceCount=0;
}*/
}
if (Confidence >= 50)
{
topConfidenceCount++;
if (topConfidenceCount == TOP_CONF_ADOPT)
{
forceNewBeat=1;
topConfidenceCount=0;
}
}
if (forceNewBeat)
{
forceNewBeat=0;
bestConfidence = Confidence;
predictionBpm=Bpm;
}
}
if (!sticked) predictionBpm = Bpm;
Bpm=predictionBpm;
/* resync = (predictionBpm &&
(predictionLastTC < TCNow - (30000/predictionBpm) - (60000/predictionBpm)*0.2) ||
(predictionLastTC < TCNow - (30000/predictionBpm) - (60000/predictionBpm)*0.2));*/
if (predictionBpm && accepted && !predicted)
{
int b;
if (TCNow > predictionLastTC + (60000 / predictionBpm)*0.7)
{
resyncin = TRUE;
b = (int)((float)predictionBpm * 1.01);
}
if (TCNow < predictionLastTC + (60000 / predictionBpm)*0.3)
{
int b;
resyncout = TRUE;
b = (int)((float)predictionBpm * 0.98);
}
if (!sticked && doResyncBpm && (resyncin || resyncout))
{
newBpm(b);
predictionBpm = GetBpm();
}
}
if (resyncin)
{
predictionLastTC = TCNow;
SliderStep(IDC_OUT, &outSlide);
doResyncBpm=TRUE;
return ((cfg_smartbeat && !cfg_smartbeatonlysticky) || (cfg_smartbeat && cfg_smartbeatonlysticky && sticked)) ? 1 : isBeat;
}
if (predicted)
{
predictionLastTC = TCNow;
if (Confidence > 25) TCHistStep(TCHist, Avg, halfDiscriminated, &hdPos, &lastTC, TCNow, BEAT_GUESSED);
SliderStep(IDC_OUT, &outSlide);
doResyncBpm=FALSE;
return ((cfg_smartbeat && !cfg_smartbeatonlysticky) || (cfg_smartbeat && cfg_smartbeatonlysticky && sticked)) ? 1 : isBeat;
}
if (resyncout)
{
predictionLastTC = TCNow;
doResyncBpm=TRUE;
return ((cfg_smartbeat && !cfg_smartbeatonlysticky) || (cfg_smartbeat && cfg_smartbeatonlysticky && sticked)) ? 0 : isBeat;
}
return ((cfg_smartbeat && !cfg_smartbeatonlysticky) || (cfg_smartbeat && cfg_smartbeatonlysticky && sticked)) ? (predictionBpm ? 0 : isBeat) : isBeat;
}
// configuration dialog stuff
BOOL CALLBACK DlgProc_Bpm(HWND hwndDlg, UINT uMsg, WPARAM wParam,LPARAM lParam)
{
switch (uMsg)
{
case WM_INITDIALOG:
inInc = 1;
outInc = 1;
inSlide = 0;
outSlide = 0;
oldDisplayBpm=-1;
oldDisplayConfidence=-1;
oldInSlide=-1;
oldOutSlide=-1;
if (cfg_smartbeat) CheckDlgButton(hwndDlg,IDC_BPMADV,BST_CHECKED); else CheckDlgButton(hwndDlg,IDC_BPMSTD,BST_CHECKED);
if (cfg_smartbeatsticky) CheckDlgButton(hwndDlg,IDC_STICKY,BST_CHECKED);
if (cfg_smartbeatresetnewsong) CheckDlgButton(hwndDlg,IDC_NEWRESET,BST_CHECKED); else CheckDlgButton(hwndDlg,IDC_NEWADAPT,BST_CHECKED);
if (cfg_smartbeatonlysticky) CheckDlgButton(hwndDlg,IDC_ONLYSTICKY,BST_CHECKED);
SendDlgItemMessage(hwndDlg, IDC_IN, TBM_SETTICFREQ, 1, 0);
SendDlgItemMessage(hwndDlg, IDC_IN, TBM_SETRANGE, TRUE, MAKELONG(0, 8));
SendDlgItemMessage(hwndDlg, IDC_OUT, TBM_SETTICFREQ, 1, 0);
SendDlgItemMessage(hwndDlg, IDC_OUT, TBM_SETRANGE, TRUE, MAKELONG(0, 8));
if (predictionBpm)
{
ShowWindow(GetDlgItem(hwndDlg, IDC_STICK), sticked ? SW_HIDE : SW_NORMAL);
ShowWindow(GetDlgItem(hwndDlg, IDC_UNSTICK), sticked ? SW_NORMAL : SW_HIDE);
}
else
{
ShowWindow(GetDlgItem(hwndDlg, IDC_STICK), SW_HIDE);
ShowWindow(GetDlgItem(hwndDlg, IDC_UNSTICK), SW_HIDE);
}
/* ShowWindow(GetDlgItem(hwndDlg, IDC_CURBPM), cfg_smartbeat ? SW_NORMAL : SW_HIDE);
ShowWindow(GetDlgItem(hwndDlg, IDC_CURCONF), cfg_smartbeat ? SW_NORMAL : SW_HIDE);
ShowWindow(GetDlgItem(hwndDlg, IDC_BPM), cfg_smartbeat ? SW_NORMAL : SW_HIDE);
ShowWindow(GetDlgItem(hwndDlg, IDC_CONFIDENCE), cfg_smartbeat ? SW_NORMAL : SW_HIDE);
ShowWindow(GetDlgItem(hwndDlg, IDC_RESET), cfg_smartbeat ? SW_NORMAL : SW_HIDE);*/
SetTimer(hwndDlg, 0, 50, NULL);
return 1;
case WM_TIMER:
{
if (oldInSlide != inSlide) {
SendDlgItemMessage(hwndDlg, IDC_IN, TBM_SETPOS, TRUE, inSlide); oldInSlide=inSlide; }
if (oldOutSlide != outSlide) {
SendDlgItemMessage(hwndDlg, IDC_OUT, TBM_SETPOS, TRUE, outSlide); oldOutSlide=outSlide; }
if (oldDisplayBpm != predictionBpm || oldsticked != sticked) {
wsprintf(txt, predictionBpm ? "%d%s"/*/%d"*/ : "Learning...", predictionBpm, cfg_smartbeatsticky && sticked ? " Got it!" : ""/*, Bpm*/);
SetDlgItemText(hwndDlg, IDC_BPM, txt);
oldDisplayBpm=predictionBpm;
oldsticked=sticked;
if (predictionBpm)
{
ShowWindow(GetDlgItem(hwndDlg, IDC_STICK), sticked ? SW_HIDE : SW_NORMAL);
ShowWindow(GetDlgItem(hwndDlg, IDC_UNSTICK), sticked ? SW_NORMAL : SW_HIDE);
}
else
{
ShowWindow(GetDlgItem(hwndDlg, IDC_STICK), SW_HIDE);
ShowWindow(GetDlgItem(hwndDlg, IDC_UNSTICK), SW_HIDE);
}
}
if (oldDisplayConfidence != Confidence) {
wsprintf(txt, "%d%%"/* (%d%%/%d%% - %d)"*/, Confidence/*, Confidence1, Confidence2, TCUsed*/);
SetDlgItemText(hwndDlg, IDC_CONFIDENCE, txt);
oldDisplayConfidence=Confidence;
}
}
return 0;
case WM_COMMAND:
if ((LOWORD(wParam) == IDC_BPMSTD) ||
(LOWORD(wParam) == IDC_BPMADV) ||
(LOWORD(wParam) == IDC_NEWRESET) ||
(LOWORD(wParam) == IDC_NEWADAPT) ||
(LOWORD(wParam) == IDC_ONLYSTICKY) ||
(LOWORD(wParam) == IDC_STICKY))
{
cfg_smartbeat=IsDlgButtonChecked(hwndDlg,IDC_BPMADV)?1:0;
cfg_smartbeatsticky=IsDlgButtonChecked(hwndDlg,IDC_STICKY)?1:0;
cfg_smartbeatresetnewsong=IsDlgButtonChecked(hwndDlg,IDC_NEWRESET)?1:0;
cfg_smartbeatonlysticky=IsDlgButtonChecked(hwndDlg,IDC_ONLYSTICKY)?1:0;
oldsticked=-1;
/* ShowWindow(GetDlgItem(hwndDlg, IDC_CURBPM), cfg_smartbeat ? SW_NORMAL : SW_HIDE);
ShowWindow(GetDlgItem(hwndDlg, IDC_CURCONF), cfg_smartbeat ? SW_NORMAL : SW_HIDE);
ShowWindow(GetDlgItem(hwndDlg, IDC_BPM), cfg_smartbeat ? SW_NORMAL : SW_HIDE);
ShowWindow(GetDlgItem(hwndDlg, IDC_CONFIDENCE), cfg_smartbeat ? SW_NORMAL : SW_HIDE);
ShowWindow(GetDlgItem(hwndDlg, IDC_RESET), cfg_smartbeat ? SW_NORMAL : SW_HIDE);*/
}
if (LOWORD(wParam) == IDC_2X)
doubleBeat();
if (LOWORD(wParam) == IDC_DIV2)
halfBeat();
if (LOWORD(wParam) == IDC_RESET)
ResetAdapt();
if (LOWORD(wParam) == IDC_STICK)
{
sticked=1;
stickyConfidenceCount=0;
}
if (LOWORD(wParam) == IDC_UNSTICK)
{
sticked=0;
stickyConfidenceCount=0;
}
return 0;
case WM_DESTROY:
KillTimer(hwndDlg, 0);
return 0;
}
return 0;
}
// Calculate BPM according to beat history
void CalcBPM(void)
{
int i;
int hdCount=0;
int r=0;
int totalTC=0, totalN=0;
float rC, etC;
int v;
double sc=0;
int mx=0;
float et;
int smSum=0, smN=0;
if (!ReadyToLearn())
return;
// First calculate average beat
for (i=0;i<TCHistSize-1;i++)
totalTC += TCHist[i].TC - TCHist[i+1].TC;
Avg = totalTC/(TCHistSize-1);
// Count how many of then are real as opposed to guessed
for (i=0;i<TCHistSize;i++)
if (TCHist[i].Type == BEAT_REAL)
r++;
// Calculate part 1 of confidence
rC = (float)min((float)((float)r / (float)TCHistSize) * 2, 1);
// Calculate typical drift
for (i=0;i<TCHistSize-1;i++)
{
v = TCHist[i].TC - TCHist[i+1].TC;
mx = max(mx, v);
sc += v*v;
}
et = (float)sqrt(sc / (TCHistSize-1) - Avg*Avg);
// Calculate confidence based on typical drift and max derivation
etC = 1 - ((float)et / (float)mx);
// Calculate confidence
Confidence = max(0, (int)(((rC * etC) * 100.0) - 50) * 2);
Confidence1 = (int)(rC * 100);
Confidence2 = (int)(etC * 100);
// Now apply second layer, recalculate average using only beats within range of typical drift
// Also, count how many of them we are keeping
totalTC=0;
for (i=0;i<TCHistSize-1;i++)
{
v += TCHist[i].TC - TCHist[i+1].TC;
if (abs(Avg-v) < et)
{
totalTC += v;
totalN++;
v = 0;
}
else
if ((float)v > Avg)
v = 0;
}
TCUsed = totalN;
// If no beat was within typical drift (how would it be possible? well lets cover our ass) then keep the simple
// average calculated earlier, else recalculate average of beats within range
if (totalN)
Avg = totalTC/totalN;
if (ReadyToGuess())
{
if (Avg) // Avg = 0 ? Ahem..
Bpm = 60000 / Avg;
if (Bpm != lastBPM)
{
newBpm(Bpm); // If realtime Bpm has changed since last time, then insert it in the smoothing tab;e
lastBPM = Bpm;
if (cfg_smartbeatsticky && predictionBpm && Confidence >= ((predictionBpm < 90) ? STICKY_THRESHOLD_LOW : STICKY_THRESHOLD))
{
stickyConfidenceCount++;
if (stickyConfidenceCount >= MIN_STICKY)
sticked=1;
}
else
stickyConfidenceCount=0;
}
Bpm = GetBpm();
// Count how many beats we discriminated
for (i=0;i<TCHistSize;i++)
if (halfDiscriminated[i]) hdCount++;
if (hdCount >= TCHistSize/2) // If we removed at least half of our beats, then we are off course. We should double our bpm
{
if (Bpm * 2 < MAX_BPM) // Lets do so only if the doubled bpm is < MAX_BPM
{
doubleBeat();
memset(halfDiscriminated, 0, TCHistSize*sizeof(int)); // Reset discrimination table
}
}
if (Bpm > 500 || Bpm < 0)
{
ResetAdapt();
}
if (Bpm < MIN_BPM)
{
if (++doubleCount > 4) // We're going too slow, lets double our bpm
doubleBeat();
}
else
doubleCount=0;
if (Bpm > MAX_BPM) // We're going too fast, lets slow our bpm by a factor of 2
{
if (++halfCount > 4)
halfBeat();
}
else
halfCount=0;
}
}
int main(int argc, char *argv[])
{
char *s;
char *labfn;
int numUtt;
void Initialise(void);
void DoGeneration(char *labfn);
if (InitShell(argc, argv, hmgens_version, hmgens_vc_id) < SUCCESS)
HError(2300, "HMGenS: InitShell failed");
InitMem();
InitMath();
InitSigP();
InitWave();
InitLabel();
InitModel();
if (InitParm() < SUCCESS)
HError(2300, "HMGenS: InitParm failed");
InitUtil();
InitFB();
InitAdapt(&xfInfo_hmm, &xfInfo_dur);
InitMap();
InitGen();
if (NumArgs() == 0)
ReportUsage();
CreateHeap(&genStack, "genStore", MSTAK, 1, 1.0, 80000, 400000);
CreateHeap(&uttStack, "uttStore", MSTAK, 1, 0.5, 100, 1000);
CreateHeap(&fbInfoStack, "FBInfoStore", MSTAK, 1, 0.5, 100, 1000);
CreateHeap(&hmmStack, "HmmStore", MSTAK, 1, 1.0, 50000, 500000);
CreateHeap(&dmStack, "dmStore", MSTAK, 1, 1.0, 50000, 500000);
SetConfParms();
CreateHMMSet(&hmset, &hmmStack, TRUE);
CreateHMMSet(&dmset, &dmStack, TRUE);
utt = (UttInfo *) New(&uttStack, sizeof(UttInfo));
genInfo = (GenInfo *) New(&genStack, sizeof(GenInfo));
fbInfo = (FBInfo *) New(&fbInfoStack, sizeof(FBInfo));
while (NextArg() == SWITCHARG) {
s = GetSwtArg();
if (strlen(s) != 1)
HError(9919, "HMGenS: Bad switch %s; must be single letter", s);
switch (s[0]) {
case 'a':
xfInfo_hmm.useInXForm = TRUE;
break;
case 'b':
xfInfo_dur.useInXForm = TRUE;
break;
case 'c':
if (NextArg() != INTARG)
HError(2119, "HMGenS: Parameter generation algorithm type value expected");
type = (ParmGenType) GetChkedInt(CHOLESKY, FB, s);
break;
case 'd':
if (NextArg() != STRINGARG)
HError(2319, "HMGenS: HMM definition directory expected");
hmmDir = GetStrArg();
break;
case 'e':
useAlign = TRUE;
break;
case 'f':
frameRate = (HTime) GetChkedFlt(0.0, 10000000.0, s);
break;
case 'g':
minFrwdP = GetChkedFlt(0.0, 1000.0, s);
break;
case 'h':
if (NextArg() != STRINGARG)
HError(1, "Speaker name pattern expected");
xfInfo_hmm.inSpkrPat = xfInfo_dur.inSpkrPat = GetStrArg();
if (NextArg() == STRINGARG)
xfInfo_hmm.paSpkrPat = xfInfo_dur.paSpkrPat = GetStrArg();
if (NextArg() != SWITCHARG)
HError(2319, "HMGenS: cannot have -h as the last option");
break;
case 'm':
modelAlign = TRUE;
break;
case 'n':
if (NextArg() != STRINGARG)
HError(2319, "HMGenS: duration model definition directory expected");
dmDir = GetStrArg();
break;
case 'p':
outPdf = TRUE;
break;
case 'r':
if (NextArg() != FLOATARG)
HError(2119, "HMGenS: Speaking rate value (float) expected");
speakRate = GetChkedFlt(0.0, 3.0, s);
break;
case 's':
stateAlign = TRUE;
break;
case 't':
pruneInit = GetChkedFlt(0.0, 1.0E20, s);
if (NextArg() == FLOATARG || NextArg() == INTARG) {
pruneInc = GetChkedFlt(0.0, 1.0E20, s);
pruneLim = GetChkedFlt(0.0, 1.0E20, s);
} else {
pruneInc = 0.0;
pruneLim = pruneInit;
}
break;
case 'v':
MSDthresh = GetChkedFlt(0.0, 1.0, s);
break;
case 'x':
if (NextArg() != STRINGARG)
HError(2319, "HMGenS: HMM file extension expected");
hmmExt = GetStrArg();
break;
case 'y':
if (NextArg() != STRINGARG)
HError(2319, "HMGenS: duration model file extension expected");
dmExt = GetStrArg();
break;
case 'B':
inBinary = TRUE;
break;
case 'E':
if (NextArg() != STRINGARG)
HError(2319, "HMGenS: parent transform directory expected");
xfInfo_hmm.usePaXForm = TRUE;
xfInfo_hmm.paXFormDir = GetStrArg();
if (NextArg() == STRINGARG)
xfInfo_hmm.paXFormExt = GetStrArg();
if (NextArg() != SWITCHARG)
HError(2319, "HMGenS: cannot have -E as the last option");
break;
case 'G':
if (NextArg() != STRINGARG)
HError(2119, "HMGenS: Label File format expected");
if ((lff = Str2Format(GetStrArg())) == ALIEN)
HError(-2189, "HMGenS: Warning ALIEN Label file format set");
break;
case 'H':
if (NextArg() != STRINGARG)
HError(3219, "HMGenS: HMM MMF File name expected");
AddMMF(&hmset, GetStrArg());
break;
case 'I':
if (NextArg() != STRINGARG)
HError(2319, "HMGenS: MLF file name expected");
LoadMasterFile(GetStrArg());
break;
case 'J':
if (NextArg() != STRINGARG)
HError(2319, "HMGenS: input transform directory expected");
AddInXFormDir(&hmset, GetStrArg());
if (NextArg() == STRINGARG)
xfInfo_hmm.inXFormExt = GetStrArg();
if (NextArg() != SWITCHARG)
HError(2319, "HMGenS: cannot have -J as the last option");
break;
case 'L':
if (NextArg() != STRINGARG)
HError(2319, "HMGenS: Label file directory expected");
labDir = GetStrArg();
break;
case 'M':
if (NextArg() != STRINGARG)
HError(2319, "HMGenS: Output macro file directory expected");
genDir = GetStrArg();
break;
case 'N':
if (NextArg() != STRINGARG)
HError(3219, "HMGenS: Duration MMF File name expected");
AddMMF(&dmset, GetStrArg());
break;
case 'T':
if (NextArg() != INTARG)
HError(2119, "HMGenS: Trace value expected");
trace = GetChkedInt(0, 0002, s);
break;
case 'W':
if (NextArg() != STRINGARG)
HError(2319, "HMGenS: parent duration transform directory expected");
xfInfo_dur.usePaXForm = TRUE;
xfInfo_dur.paXFormDir = GetStrArg();
if (NextArg() == STRINGARG)
xfInfo_dur.paXFormExt = GetStrArg();
if (NextArg() != SWITCHARG)
HError(2319, "HMGenS: cannot have -W as the last option");
break;
case 'X':
if (NextArg() != STRINGARG)
HError(2319, "HMGenS: Label file extension expected");
labExt = GetStrArg();
break;
case 'Y':
if (NextArg() != STRINGARG)
HError(2319, "HMGenS: input duration transform directory expected");
AddInXFormDir(&dmset, GetStrArg());
if (NextArg() == STRINGARG) {
if (xfInfo_dur.inXFormExt == NULL)
xfInfo_dur.inXFormExt = GetStrArg();
else
HError(2319, "MGenS: only one input duration transform extension may be specified");
}
if (NextArg() != SWITCHARG)
HError(2319, "HMGenS: cannot have -Y as the last option");
break;
default:
HError(9919, "HMGenS: Unknown switch %s", s);
}
}
if (NextArg() != STRINGARG)
HError(2319, "HMGenS: file name of vocabulary list expected");
Initialise();
InitUttInfo(utt, FALSE);
numUtt = 1;
if (trace & T_TOP)
SetTraceGen();
/* generate parameter sequences */
do {
if (NextArg() != STRINGARG)
HError(2319, "HMGenS: data file name expected");
labfn = GetStrArg();
/* track speakers */
if (UpdateSpkrStats(&hmset, &xfInfo_hmm, labfn)) {
if (!xfInfo_hmm.useInXForm)
xfInfo_hmm.inXForm = NULL;
}
if (UpdateSpkrStats(&dmset, &xfInfo_dur, labfn)) {
if (!xfInfo_dur.useInXForm)
xfInfo_dur.inXForm = NULL;
else
ResetDMMPreComps(&dmset);
}
fbInfo->xfinfo_hmm = &xfInfo_hmm;
fbInfo->xfinfo_dur = &xfInfo_dur;
fbInfo->inXForm_hmm = xfInfo_hmm.inXForm;
fbInfo->inXForm_dur = xfInfo_dur.inXForm;
fbInfo->al_inXForm_hmm = xfInfo_hmm.al_inXForm;
fbInfo->al_inXForm_dur = xfInfo_dur.al_inXForm;
fbInfo->paXForm_hmm = xfInfo_hmm.paXForm;
fbInfo->paXForm_dur = xfInfo_dur.paXForm;
/* generate parameters */
DoGeneration(labfn);
numUtt++;
} while (NumArgs() > 0);
if (trace & T_TOP) {
printf("Generation complete - average log prob per frame = %e (%d frames)\n", totalPr / totalT, totalT);
}
/* Reset stacks */
Dispose(&fbInfoStack, fbInfo);
Dispose(&genStack, genInfo);
Dispose(&uttStack, utt);
ResetHeap(&fbInfoStack);
ResetHeap(&uttStack);
ResetHeap(&genStack);
ResetHeap(&dmStack);
ResetHeap(&hmmStack);
/* Reset modules */
ResetGen();
ResetAdapt(&xfInfo_hmm, &xfInfo_dur);
ResetFB();
ResetUtil();
ResetParm();
ResetModel();
ResetLabel();
ResetWave();
ResetSigP();
ResetMath();
ResetMem();
ResetShell();
Exit(0);
return (0); /* never reached -- make compiler happy */
}
