MODEL prev_model; /* previous frame's model parameters */
float prev_lsps_[LPC_ORD]; /* previous frame's LSPs */
float prev_energy; /* previous frame's LPC energy */
+
+ float xq_enc[2]; /* joint pitch and energy VQ states */
+ float xq_dec[2];
};
/*---------------------------------------------------------------------------*\
void codec2_decode_2500(struct CODEC2 *c2, short speech[], const unsigned char * bits);
void codec2_encode_1500(struct CODEC2 *c2, unsigned char * bits, short speech[]);
void codec2_decode_1500(struct CODEC2 *c2, short speech[], const unsigned char * bits);
+void codec2_encode_1400(struct CODEC2 *c2, unsigned char * bits, short speech[]);
+void codec2_decode_1400(struct CODEC2 *c2, short speech[], const unsigned char * bits);
void codec2_encode_1200(struct CODEC2 *c2, unsigned char * bits, short speech[]);
void codec2_decode_1200(struct CODEC2 *c2, short speech[], const unsigned char * bits);
assert(
(mode == CODEC2_MODE_2500) ||
(mode == CODEC2_MODE_1500) ||
+ (mode == CODEC2_MODE_1400) ||
(mode == CODEC2_MODE_1200)
);
c2->mode = mode;
return NULL;
}
+ c2->xq_enc[0] = c2->xq_enc[1] = 0.0;
+ c2->xq_dec[0] = c2->xq_dec[1] = 0.0;
+
return c2;
}
return 50;
if (c2->mode == CODEC2_MODE_1500)
return 60;
+ if (c2->mode == CODEC2_MODE_1400)
+ return 56;
if (c2->mode == CODEC2_MODE_1200)
return 48;
return 160;
if (c2->mode == CODEC2_MODE_1500)
return 320;
+ if (c2->mode == CODEC2_MODE_1400)
+ return 320;
if (c2->mode == CODEC2_MODE_1200)
return 320;
assert(
(c2->mode == CODEC2_MODE_2500) ||
(c2->mode == CODEC2_MODE_1500) ||
+ (c2->mode == CODEC2_MODE_1400) ||
(c2->mode == CODEC2_MODE_1200)
);
codec2_encode_2500(c2, bits, speech);
if (c2->mode == CODEC2_MODE_1500)
codec2_encode_1500(c2, bits, speech);
+ if (c2->mode == CODEC2_MODE_1400)
+ codec2_encode_1400(c2, bits, speech);
if (c2->mode == CODEC2_MODE_1200)
codec2_encode_1200(c2, bits, speech);
}
assert(
(c2->mode == CODEC2_MODE_2500) ||
(c2->mode == CODEC2_MODE_1500) ||
+ (c2->mode == CODEC2_MODE_1400) ||
(c2->mode == CODEC2_MODE_1200)
);
codec2_decode_2500(c2, speech, bits);
if (c2->mode == CODEC2_MODE_1500)
codec2_decode_1500(c2, speech, bits);
+ if (c2->mode == CODEC2_MODE_1400)
+ codec2_decode_1400(c2, speech, bits);
if (c2->mode == CODEC2_MODE_1200)
codec2_decode_1200(c2, speech, bits);
}
int i;
unsigned int nbit = 0;
MODEL model_interp;
- static int frames;
+ //static int frames;
//fprintf(stderr,"frame: %d\n", frames+=2);
assert(c2 != NULL);
AUTHOR......: David Rowe
DATE CREATED: Nov 14 2011
- Encodes 320 speech samples (40ms of speech) into 60 bits.
+ Encodes 320 speech samples (40ms of speech) into 60 bits.
The codec2 algorithm actually operates internally on 10ms (80
sample) frames, so we run the encoding algorithm for times:
frame 0: just send voicing bit
- frame 1: scalar quantisation of LSPs and Wo and E
+ frame 1: scalar quantisation of LSPs, Wo and E
frame 2: just send voicing bit
- frame 3: delta-time quantisation Wo and E
+ frame 3: delta-time Wo and scalar E
The bit allocation is:
FUNCTION....: codec2_decode_1500
AUTHOR......: David Rowe
- DATE CREATED: 16 Nov 2011
+ DATE CREATED: 11 May 2012
- Decodes frames of 56 bits into 320 samples (40ms) of speech.
+ Decodes frames of 60 bits into 320 samples (40ms) of speech.
\*---------------------------------------------------------------------------*/
}
+
+/*---------------------------------------------------------------------------*\
+
+ FUNCTION....: codec2_encode_1400
+ AUTHOR......: David Rowe
+ DATE CREATED: May 11 2012
+
+ Encodes 320 speech samples (40ms of speech) into 56 bits.
+
+ The codec2 algorithm actually operates internally on 10ms (80
+ sample) frames, so we run the encoding algorithm for times:
+
+ frame 0: just send voicing bit
+ frame 1: scalar quantisation of LSPs, joint VQ of Wo and E
+ frame 2: just send voicing bit
+ frame 3: joint VQ of Wo and E
+
+ The bit allocation is:
+
+ Parameter frame 2 frame 4 Total
+ -------------------------------------------------------
+ Harmonic magnitudes (LSPs) 36 0 36
+ Energy+Wo 8 8 16
+ Voicing (10ms update) 2 2 4
+ TOTAL 46 10 56
+
+\*---------------------------------------------------------------------------*/
+
+void codec2_encode_1400(struct CODEC2 *c2, unsigned char * bits, short speech[])
+{
+ MODEL model;
+ float lsps[LPC_ORD], lsps_[LPC_ORD];
+ float ak[LPC_ORD+1];
+ float e;
+ int voiced1, voiced2, voiced3, voiced4;
+ int lsp_indexes[LPC_ORD];
+ int WoE_index;
+ int i;
+ unsigned int nbit = 0;
+ unsigned int nbit_tmp;
+ //static int frames;
+
+ assert(c2 != NULL);
+
+ memset(bits, '\0', ((codec2_bits_per_frame(c2) + 7) / 8));
+
+ /* frame 1: - we just want voicing -------------------------------- */
+
+ //fprintf(stderr,"frame: %d\n", ++frames);
+ analyse_one_frame(c2, &model, speech);
+ voiced1 = model.voiced;
+
+ /* frame 2: - full LSP, joint Wo & EE ----------------------------- */
+
+ //fprintf(stderr,"frame: %d\n", ++frames);
+ analyse_one_frame(c2, &model, &speech[N]);
+ voiced2 = model.voiced;
+
+ e = speech_to_uq_lsps(lsps, ak, c2->Sn, c2->w, LPC_ORD);
+ encode_lsps_scalar(lsp_indexes, lsps, LPC_ORD);
+ WoE_index = encode_WoE(&model, e, c2->xq_enc);
+
+ for(i=0; i<LSP_SCALAR_INDEXES; i++) {
+ pack(bits, &nbit, lsp_indexes[i], lsp_bits(i));
+ }
+ pack(bits, &nbit, WoE_index, WO_E_BITS);
+ pack(bits, &nbit, voiced1, 1);
+ pack(bits, &nbit, voiced2, 1);
+
+ /* decode LSPs for testing */
+
+ decode_lsps_scalar(lsps_, lsp_indexes, LPC_ORD);
+ bw_expand_lsps(lsps_, LPC_ORD);
+ /*
+ fprintf(stderr,"\n lsps_......: ");
+ for(i=0; i<LPC_ORD; i++)
+ fprintf(stderr,"%5.3f ", lsps_[i]);
+ fprintf(stderr,"\n");
+ */
+
+ /* frame 3: - we just want voicing --------------------------------- */
+
+ //fprintf(stderr,"frame: %d\n", ++frames);
+ analyse_one_frame(c2, &model, &speech[2*N]);
+ voiced3 = model.voiced;
+
+ /* frame 4: - voicing and joint Wo & E ---------------------------- */
+
+ //fprintf(stderr,"frame: %d\n", ++frames);
+ analyse_one_frame(c2, &model, &speech[3*N]);
+ voiced4 = model.voiced;
+
+ /* need to run this to get LPC energy */
+ e = speech_to_uq_lsps(lsps, ak, c2->Sn, c2->w, LPC_ORD);
+
+ WoE_index = encode_WoE(&model, e, c2->xq_enc);
+
+ //fprintf(stderr," e: %f code: %d dec: %f \n", e, energy_index, decode_energy(energy_index));
+
+ nbit_tmp = nbit;
+ pack(bits, &nbit, WoE_index, WO_E_BITS);
+ pack(bits, &nbit, voiced3, 1);
+ pack(bits, &nbit, voiced4, 1);
+ //fprintf(stderr," 00 16 24 32 40 48 56\n");
+ //fprintf(stderr,"nbit = %d %02x %02x %02x %02x %02x %02x %02x %02x\n", nbit,
+ // bits[0], bits[1], bits[2], bits[3],
+ // bits[4], bits[5], bits[6], bits[7]);
+
+ //fprintf(stderr," nbit_tmp: %d ", nbit_tmp);
+ // fprintf(stderr,"energy_index after: %d\n", energy_index);
+
+ assert(nbit == (unsigned)codec2_bits_per_frame(c2));
+ //if (frames == 36)
+ //exit(0);
+}
+
+
+/*---------------------------------------------------------------------------*\
+
+ FUNCTION....: codec2_decode_1400
+ AUTHOR......: David Rowe
+ DATE CREATED: 11 May 2012
+
+ Decodes frames of 48 bits into 320 samples (40ms) of speech.
+
+\*---------------------------------------------------------------------------*/
+
+void codec2_decode_1400(struct CODEC2 *c2, short speech[], const unsigned char * bits)
+{
+ MODEL model;
+ int voiced1, voiced2, voiced3, voiced4;
+ int lsp_indexes[LPC_ORD];
+ float lsps_[LPC_ORD];
+ int WoE_index;
+ float energy;
+ float snr;
+ float ak[LPC_ORD+1];
+ float ak_interp[LPC_ORD+1];
+ float lsps_interp[LPC_ORD];
+ int i;
+ unsigned int nbit = 0;
+ MODEL model_interp;
+ static int frames;
+ float prev__Wo;
+
+ assert(c2 != NULL);
+
+ /* unpack frame 1 & 2 bit stream to integer codes */
+
+ for(i=0; i<LSP_SCALAR_INDEXES; i++) {
+ lsp_indexes[i] = unpack(bits, &nbit, lsp_bits(i));
+ }
+ WoE_index = unpack(bits, &nbit, WO_E_BITS);
+ voiced1 = unpack(bits, &nbit, 1);
+ voiced2 = unpack(bits, &nbit, 1);
+
+ /* decode codes to model parameters */
+
+ decode_WoE(&model, &energy, c2->xq_dec, WoE_index);
+ memset(&model.A, 0, (model.L+1)*sizeof(model.A[0]));
+
+ /* decode frame 2 LSPs and model amplitudes */
+
+ decode_lsps_scalar(lsps_, lsp_indexes, LPC_ORD);
+ check_lsp_order(lsps_, LPC_ORD);
+ bw_expand_lsps(lsps_, LPC_ORD);
+ lsp_to_lpc(lsps_, ak, LPC_ORD);
+ aks_to_M2(ak, LPC_ORD, &model, energy, &snr, 1);
+ apply_lpc_correction(&model);
+
+ /* interpolate frame 1 model parameters from adjacent frames */
+
+ model.voiced = voiced2;
+ model_interp.voiced = voiced1;
+ model_interp.Wo = P_MAX/2;
+ memset(&model_interp.A, 0, MAX_AMP*sizeof(model_interp.A[0]));
+
+ interpolate_lsp(&model_interp, &c2->prev_model, &model,
+ c2->prev_lsps_, c2->prev_energy, lsps_, energy, ak_interp,
+ lsps_interp);
+ apply_lpc_correction(&model_interp);
+
+ frames += 2;
+ /* used for comparing to c2sim version
+ fprintf(stderr,"frame: %d\n", frames);
+ fprintf(stderr," Wo: %1.5f L: %d v1: %d prev_e: %f\n",
+ model_interp.Wo, model_interp.L, model_interp.voiced, c2->prev_energy);
+ fprintf(stderr," lsps_interp: ");
+ for(i=0; i<LPC_ORD; i++)
+ fprintf(stderr,"%5.3f ", lsps_interp[i]);
+ fprintf(stderr,"\n A..........: ");
+ for(i=0; i<10; i++)
+ fprintf(stderr,"%5.3f ",model_interp.A[i]);
+
+ fprintf(stderr,"\n Wo: %1.5f L: %d e: %3.2f v2: %d\n",
+ model.Wo, model.L, energy, model.voiced);
+ fprintf(stderr," lsps_......: ");
+ for(i=0; i<LPC_ORD; i++)
+ fprintf(stderr,"%5.3f ", lsps_[i]);
+ fprintf(stderr,"\n A..........: ");
+ for(i=0; i<10; i++)
+ fprintf(stderr,"%5.3f ",model.A[i]);
+ fprintf(stderr,"\n");
+ */
+
+ /* synthesise frame 1 and frame 2 10ms frames */
+
+ synthesise_one_frame(c2, speech, &model_interp, ak_interp);
+ //fprintf(stderr," buf[0] %d\n", speech[0]);
+ synthesise_one_frame(c2, &speech[N], &model, ak);
+ //fprintf(stderr," buf[0] %d\n", speech[N]);
+
+ /* update memories (decode states) for next time */
+
+ memcpy(&c2->prev_model, &model, sizeof(MODEL));
+ memcpy(c2->prev_lsps_, lsps_, sizeof(lsps_));
+ c2->prev_energy = energy;
+ prev__Wo = model.Wo;
+
+ /*--------------------------------------------------------------------*/
+
+ /* unpack frame 3 & 4 bit stream to integer codes */
+
+ WoE_index = unpack(bits, &nbit, WO_E_BITS);
+ voiced3 = unpack(bits, &nbit, 1);
+ voiced4 = unpack(bits, &nbit, 1);
+ assert(nbit == (unsigned)codec2_bits_per_frame(c2));
+
+ /* decode integer codes to model parameters */
+
+ memset(&model.A, 0, (model.L+1)*sizeof(model.A[0]));
+
+ /* decode frame 4 */
+
+ decode_WoE(&model, &energy, c2->xq_dec, WoE_index);
+ aks_to_M2(ak, LPC_ORD, &model, energy, &snr, 1);
+ apply_lpc_correction(&model);
+
+ /* interpolate frame 3 model parameters from adjacent frames */
+
+ model.voiced = voiced4;
+ model_interp.voiced = voiced3;
+ model_interp.Wo = P_MAX/2;
+ memset(&model_interp.A, 0, MAX_AMP*sizeof(model_interp.A[0]));
+
+ interpolate_lsp(&model_interp, &c2->prev_model, &model,
+ c2->prev_lsps_, c2->prev_energy, lsps_, energy, ak_interp,
+ lsps_interp);
+ apply_lpc_correction(&model_interp);
+
+ frames +=2;
+ /* used for comparing to c2sim version:
+ fprintf(stderr,"frame: %d\n", frames);
+
+ fprintf(stderr," Wo: %1.5f L: %d v1: %d prev_e: %f\n",
+ model_interp.Wo, model_interp.L, model_interp.voiced, c2->prev_energy);
+ fprintf(stderr," lsps_interp: ");
+ for(i=0; i<LPC_ORD; i++)
+ fprintf(stderr,"%5.3f ", lsps_interp[i]);
+ fprintf(stderr,"\n A..........: ");
+ for(i=0; i<10; i++)
+ fprintf(stderr,"%5.3f ",model_interp.A[i]);
+
+ fprintf(stderr,"\n Wo: %1.5f L: %d e_index: %d e: %3.2f v2: %d\n",
+ model.Wo, model.L, energy_index, energy, model.voiced);
+ fprintf(stderr," lsps_......: ");
+ for(i=0; i<LPC_ORD; i++)
+ fprintf(stderr,"%5.3f ", lsps_[i]);
+ fprintf(stderr,"\n A..........: ");
+ for(i=0; i<10; i++)
+ fprintf(stderr,"%5.3f ",model.A[i]);
+ fprintf(stderr,"\n");
+ */
+
+ /* synthesise frame 3 and frame 4 10ms frames */
+
+ synthesise_one_frame(c2, &speech[2*N], &model_interp, ak_interp);
+ //fprintf(stderr," buf[0] %d\n", speech[2*N]);
+ synthesise_one_frame(c2, &speech[3*N], &model, ak);
+ //fprintf(stderr," buf[0] %d\n", speech[3*N]);
+
+ if (frames == 44) {
+ //exit(0);
+ }
+
+ /* update memories (decode states) for next time */
+
+ memcpy(&c2->prev_model, &model, sizeof(MODEL));
+ memcpy(c2->prev_lsps_, lsps_, sizeof(lsps_));
+ c2->prev_energy = energy;
+
+}
+
+
/*---------------------------------------------------------------------------*\
FUNCTION....: codec2_encode_1200
/*---------------------------------------------------------------------------*\
- FUNCTION....: quantize_ge()
+ FUNCTION....: quantise_WoE()
AUTHOR......: Jean-Marc Valin & David Rowe
DATE CREATED: 29 Feb 2012
- Experimental joint Wo and LPC energy vector quantiser developed my
- Jean-Marc Valin.
+ Experimental joint Wo and LPC energy vector quantiser developed by
+ Jean-Marc Valin. Exploits correlations between the difference in
+ the log pitch and log energy from frame to frame. For example the
+ both the pitch and energy tend to only change by small amounts
+ during voiced speech, however it is important that these changes be
+ coded carefully. During unvoiced speech they both change a lot but
+ the ear is less sensitve to errors so coarser quantisation is OK.
+
+ The ear is sensitive to log energy and loq pitch so we quantise in
+ thise domains. That way the error measure used to quantise the
+ values is close to way the ear senses errors.
+
+ See http://jmspeex.livejournal.com/10446.html
+
+\*---------------------------------------------------------------------------*/
+
+void quantise_WoE(MODEL *model, float *e, float xq[])
+{
+ int i, n1;
+ float x[2];
+ float err[2];
+ float w[2];
+ const float *codebook1 = ge_cb[0].cb;
+ int nb_entries = ge_cb[0].m;
+ int ndim = ge_cb[0].k;
+ float Wo_min = TWO_PI/P_MAX;
+ float Wo_max = TWO_PI/P_MIN;
+
+ x[0] = log10((model->Wo/PI)*4000.0/50.0)/log10(2);
+ x[1] = 10.0*log10(1e-4 + *e);
+
+ compute_weights2(x, xq, w, ndim);
+ for (i=0;i<ndim;i++)
+ err[i] = x[i]-ge_coeff[i]*xq[i];
+ n1 = find_nearest_weighted(codebook1, nb_entries, err, w, ndim);
+
+ for (i=0;i<ndim;i++)
+ {
+ xq[i] = ge_coeff[i]*xq[i] + codebook1[ndim*n1+i];
+ err[i] -= codebook1[ndim*n1+i];
+ }
+
+ /*
+ x = log2(4000*Wo/(PI*50));
+ 2^x = 4000*Wo/(PI*50)
+ Wo = (2^x)*(PI*50)/4000;
+ */
+
+ model->Wo = pow(2.0, xq[0])*(PI*50.0)/4000.0;
+
+ /* bit errors can make us go out of range leading to all sorts of
+ probs like seg faults */
+
+ if (model->Wo > Wo_max) model->Wo = Wo_max;
+ if (model->Wo < Wo_min) model->Wo = Wo_min;
+
+ model->L = PI/model->Wo; /* if we quantise Wo re-compute L */
+
+ *e = pow(10.0, xq[1]/10.0);
+}
+
+/*---------------------------------------------------------------------------*\
+
+ FUNCTION....: encode_WoE()
+ AUTHOR......: Jean-Marc Valin & David Rowe
+ DATE CREATED: 11 May 2012
+
+ Joint Wo and LPC energy vector quantiser developed my Jean-Marc
+ Valin. Returns index, and updated states xq[].
\*---------------------------------------------------------------------------*/
-void ge_quantise(float *x, float *xq)
+int encode_WoE(MODEL *model, float e, float xq[])
{
int i, n1;
+ float x[2];
float err[2];
float w[2];
const float *codebook1 = ge_cb[0].cb;
int nb_entries = ge_cb[0].m;
int ndim = ge_cb[0].k;
- //printf("ndim %d nb_entries %d\n", ndim, nb_entries);
+ assert((1<<WO_E_BITS) == nb_entries);
+
+ x[0] = log10((model->Wo/PI)*4000.0/50.0)/log10(2);
+ x[1] = 10.0*log10(1e-4 + e);
+
compute_weights2(x, xq, w, ndim);
- //exit(0);
for (i=0;i<ndim;i++)
err[i] = x[i]-ge_coeff[i]*xq[i];
n1 = find_nearest_weighted(codebook1, nb_entries, err, w, ndim);
err[i] -= codebook1[ndim*n1+i];
}
+ return n1;
+}
+
+
+/*---------------------------------------------------------------------------*\
+
+ FUNCTION....: decode_WoE()
+ AUTHOR......: Jean-Marc Valin & David Rowe
+ DATE CREATED: 11 May 2012
+
+ Joint Wo and LPC energy vector quantiser developed my Jean-Marc
+ Valin. Given index and states xq[], returns Wo & E, and updates
+ states xq[].
+
+\*---------------------------------------------------------------------------*/
+
+void decode_WoE(MODEL *model, float *e, float xq[], int n1)
+{
+ int i;
+ float err[2];
+ const float *codebook1 = ge_cb[0].cb;
+ int ndim = ge_cb[0].k;
+ float Wo_min = TWO_PI/P_MAX;
+ float Wo_max = TWO_PI/P_MIN;
+
+ for (i=0;i<ndim;i++)
+ {
+ xq[i] = ge_coeff[i]*xq[i] + codebook1[ndim*n1+i];
+ err[i] -= codebook1[ndim*n1+i];
+ }
+
+ model->Wo = pow(2.0, xq[0])*(PI*50.0)/4000.0;
+
+ /* bit errors can make us go out of range leading to all sorts of
+ probs like seg faults */
+
+ if (model->Wo > Wo_max) model->Wo = Wo_max;
+ if (model->Wo < Wo_min) model->Wo = Wo_min;
+
+ model->L = PI/model->Wo; /* if we quantise Wo re-compute L */
+
+ *e = pow(10.0, xq[1]/10.0);
}
projects / freetel-svn-tracking.git / commitdiff