VorbisPsy *vorbis_psy_init(int rate, int n)
{
long i,j,lo=-99,hi=1;
VorbisPsy *p = speex_alloc(sizeof(VorbisPsy));
memset(p,0,sizeof(*p));
p->n = n;
spx_drft_init(&p->lookup, n);
p->bark = speex_alloc(n*sizeof(*p->bark));
p->rate=rate;
p->vi = &example_tuning;
/* BH4 window */
p->window = speex_alloc(sizeof(*p->window)*n);
float a0 = .35875f;
float a1 = .48829f;
float a2 = .14128f;
float a3 = .01168f;
for(i=0;i<n;i++)
p->window[i] = //a0 - a1*cos(2.*M_PI/n*(i+.5)) + a2*cos(4.*M_PI/n*(i+.5)) - a3*cos(6.*M_PI/n*(i+.5));
sin((i+.5)/n * M_PI)*sin((i+.5)/n * M_PI);
/* bark scale lookups */
for(i=0;i<n;i++){
float bark=toBARK(rate/(2*n)*i);
for(;lo+p->vi->noisewindowlomin<i &&
toBARK(rate/(2*n)*lo)<(bark-p->vi->noisewindowlo);lo++);
for(;hi<=n && (hi<i+p->vi->noisewindowhimin ||
toBARK(rate/(2*n)*hi)<(bark+p->vi->noisewindowhi));hi++);
p->bark[i]=((lo-1)<<16)+(hi-1);
}
/* set up rolling noise median */
p->noiseoffset=speex_alloc(n*sizeof(*p->noiseoffset));
for(i=0;i<n;i++){
float halfoc=toOC((i+.5)*rate/(2.*n))*2.;
int inthalfoc;
float del;
if(halfoc<0)halfoc=0;
if(halfoc>=P_BANDS-1)halfoc=P_BANDS-1;
inthalfoc=(int)halfoc;
del=halfoc-inthalfoc;
p->noiseoffset[i]=
p->vi->noiseoff[inthalfoc]*(1.-del) +
p->vi->noiseoff[inthalfoc+1]*del;
}
#if 0
_analysis_output_always("noiseoff0",ls,p->noiseoffset,n,1,0,0);
#endif
return p;
}
/** Creates a new echo canceller state */
SpeexEchoState *speex_echo_state_init(int frame_size, int filter_length)
{
int i,j,N,M;
SpeexEchoState *st = (SpeexEchoState *)speex_alloc(sizeof(SpeexEchoState));
st->frame_size = frame_size;
st->window_size = 2*frame_size;
N = st->window_size;
M = st->M = (filter_length+st->frame_size-1)/frame_size;
st->cancel_count=0;
st->adapt_rate = .01f;
st->sum_adapt = 0;
st->Sey = 0;
st->Syy = 0;
st->See = 0;
st->fft_lookup = (struct drft_lookup*)speex_alloc(sizeof(struct drft_lookup));
spx_drft_init(st->fft_lookup, N);
st->x = (float*)speex_alloc(N*sizeof(float));
st->d = (float*)speex_alloc(N*sizeof(float));
st->y = (float*)speex_alloc(N*sizeof(float));
st->y2 = (float*)speex_alloc(N*sizeof(float));
st->Yps = (float*)speex_alloc(N*sizeof(float));
st->last_y = (float*)speex_alloc(N*sizeof(float));
st->Yf = (float*)speex_alloc((st->frame_size+1)*sizeof(float));
st->Rf = (float*)speex_alloc((st->frame_size+1)*sizeof(float));
st->Xf = (float*)speex_alloc((st->frame_size+1)*sizeof(float));
st->fratio = (float*)speex_alloc((st->frame_size+1)*sizeof(float));
st->regul = (float*)speex_alloc(N*sizeof(float));
st->X = (float*)speex_alloc(M*N*sizeof(float));
st->D = (float*)speex_alloc(N*sizeof(float));
st->Y = (float*)speex_alloc(N*sizeof(float));
st->Y2 = (float*)speex_alloc(N*sizeof(float));
st->E = (float*)speex_alloc(N*sizeof(float));
st->W = (float*)speex_alloc(M*N*sizeof(float));
st->PHI = (float*)speex_alloc(M*N*sizeof(float));
st->power = (float*)speex_alloc((frame_size+1)*sizeof(float));
st->power_1 = (float*)speex_alloc((frame_size+1)*sizeof(float));
st->grad = (float*)speex_alloc(N*M*sizeof(float));
for (i=0; i<N*M; i++)
{
st->W[i] = st->PHI[i] = 0;
}
st->regul[0] = (.01+(10.)/((4.)*(4.)))/M;
for (i=1,j=1; i<N-1; i+=2,j++)
{
st->regul[i] = .01+((10.)/((j+4.)*(j+4.)))/M;
st->regul[i+1] = .01+((10.)/((j+4.)*(j+4.)))/M;
}
st->regul[i] = .01+((10.)/((j+4.)*(j+4.)))/M;
st->adapted = 0;
return st;
}
EXPORT SpeexDecorrState *speex_decorrelate_new(int rate, int channels,
int frame_size)
{
int i, ch;
SpeexDecorrState *st = speex_alloc(sizeof(SpeexDecorrState));
st->rate = rate;
st->channels = channels;
st->frame_size = frame_size;
#ifdef VORBIS_PSYCHO
st->psy = vorbis_psy_init(rate, 2 * frame_size);
spx_drft_init(&st->lookup, 2 * frame_size);
st->wola_mem = speex_alloc(frame_size * sizeof(float));
st->curve = speex_alloc(frame_size * sizeof(float));
#endif
st->y = speex_alloc(frame_size * sizeof(float));
st->buff = speex_alloc(channels * 2 * frame_size * sizeof(float));
st->ringID = speex_alloc(channels * sizeof(int));
st->order = speex_alloc(channels * sizeof(int));
st->alpha = speex_alloc(channels * sizeof(float));
st->ring = speex_alloc(channels * ALLPASS_ORDER * sizeof(float));
/*FIXME: The +20 is there only as a kludge for ALL_PASS_OLA */
st->vorbis_win = speex_alloc((2 * frame_size + 20) * sizeof(float));
for (i = 0; i < 2 * frame_size; i++)
st->vorbis_win[i] =
sin(.5 * M_PI * sin(M_PI * i / (2 * frame_size)) *
sin(M_PI * i / (2 * frame_size)));
st->seed = rand();
for (ch = 0; ch < channels; ch++) {
for (i = 0; i < ALLPASS_ORDER; i++)
st->ring[ch][i] = 0;
st->ringID[ch] = 0;
st->alpha[ch] = 0;
st->order[ch] = 10;
}
return st;
}
SpeexPreprocessState *speex_preprocess_state_init(int frame_size, int sampling_rate)
{
int i;
int N, N3, N4;
SpeexPreprocessState *st = (SpeexPreprocessState *)speex_alloc(sizeof(SpeexPreprocessState));
st->frame_size = frame_size;
/* Round ps_size down to the nearest power of two */
#if 0
i=1;
st->ps_size = st->frame_size;
while(1)
{
if (st->ps_size & ~i)
{
st->ps_size &= ~i;
i<<=1;
} else {
break;
}
}
if (st->ps_size < 3*st->frame_size/4)
st->ps_size = st->ps_size * 3 / 2;
#else
st->ps_size = st->frame_size;
#endif
N = st->ps_size;
N3 = 2*N - st->frame_size;
N4 = st->frame_size - N3;
st->sampling_rate = sampling_rate;
st->denoise_enabled = 1;
st->agc_enabled = 0;
st->agc_level = 8000;
st->vad_enabled = 0;
st->dereverb_enabled = 0;
st->reverb_decay = .5;
st->reverb_level = .2;
st->frame = (float*)speex_alloc(2*N*sizeof(float));
st->ps = (float*)speex_alloc(N*sizeof(float));
st->gain2 = (float*)speex_alloc(N*sizeof(float));
st->window = (float*)speex_alloc(2*N*sizeof(float));
st->noise = (float*)speex_alloc(N*sizeof(float));
st->reverb_estimate = (float*)speex_alloc(N*sizeof(float));
st->old_ps = (float*)speex_alloc(N*sizeof(float));
st->gain = (float*)speex_alloc(N*sizeof(float));
st->prior = (float*)speex_alloc(N*sizeof(float));
st->post = (float*)speex_alloc(N*sizeof(float));
st->loudness_weight = (float*)speex_alloc(N*sizeof(float));
st->inbuf = (float*)speex_alloc(N3*sizeof(float));
st->outbuf = (float*)speex_alloc(N3*sizeof(float));
st->echo_noise = (float*)speex_alloc(N*sizeof(float));
st->S = (float*)speex_alloc(N*sizeof(float));
st->Smin = (float*)speex_alloc(N*sizeof(float));
st->Stmp = (float*)speex_alloc(N*sizeof(float));
st->update_prob = (float*)speex_alloc(N*sizeof(float));
st->zeta = (float*)speex_alloc(N*sizeof(float));
st->Zpeak = 0;
st->Zlast = 0;
st->noise_bands = (float*)speex_alloc(NB_BANDS*sizeof(float));
st->noise_bands2 = (float*)speex_alloc(NB_BANDS*sizeof(float));
st->speech_bands = (float*)speex_alloc(NB_BANDS*sizeof(float));
st->speech_bands2 = (float*)speex_alloc(NB_BANDS*sizeof(float));
st->noise_bandsN = st->speech_bandsN = 1;
conj_window(st->window, 2*N3);
for (i=2*N3;i<2*st->ps_size;i++)
st->window[i]=1;
if (N4>0)
{
for (i=N3-1;i>=0;i--)
{
st->window[i+N3+N4]=st->window[i+N3];
st->window[i+N3]=1;
}
}
for (i=0;i<N;i++)
{
st->noise[i]=1e4;
st->reverb_estimate[i]=0.;
st->old_ps[i]=1e4;
st->gain[i]=1;
st->post[i]=1;
st->prior[i]=1;
}
for (i=0;i<N3;i++)
{
st->inbuf[i]=0;
st->outbuf[i]=0;
}
for (i=0;i<N;i++)
{
float ff=((float)i)*.5*sampling_rate/((float)N);
st->loudness_weight[i] = .35f-.35f*ff/16000.f+.73f*exp(-.5f*(ff-3800)*(ff-3800)/9e5f);
if (st->loudness_weight[i]<.01f)
st->loudness_weight[i]=.01f;
st->loudness_weight[i] *= st->loudness_weight[i];
}
st->speech_prob = 0;
st->last_speech = 1000;
st->loudness = pow(6000,LOUDNESS_EXP);
st->loudness2 = 6000;
st->nb_loudness_adapt = 0;
st->fft_lookup = (struct drft_lookup*)speex_alloc(sizeof(struct drft_lookup));
spx_drft_init(st->fft_lookup,2*N);
st->nb_adapt=0;
st->consec_noise=0;
st->nb_preprocess=0;
return st;
}
