int main(int argc, char *argv[])
{
struct CODEC2 *codec2;
FILE *fin;
FILE *fout;
short *buf;
unsigned char *bits;
int nsam, nbit, i, r;
for(i=0; i<10; i++) {
r = codec2_rand();
printf("[%d] r = %d\n", i, r);
}
if (argc != 3) {
printf("usage: %s InputRawSpeechFile OutputRawSpeechFile\n", argv[0]);
exit(1);
}
if ( (fin = fopen(argv[1],"rb")) == NULL ) {
fprintf(stderr, "Error opening input speech file: %s: %s.\n",
argv[1], strerror(errno));
exit(1);
}
if ( (fout = fopen(argv[2],"wb")) == NULL ) {
fprintf(stderr, "Error opening output speech file: %s: %s.\n",
argv[2], strerror(errno));
exit(1);
}
#ifdef DUMP
dump_on("c2demo");
#endif
/* Note only one set of Codec 2 states is required for an encoder
and decoder pair. */
codec2 = codec2_create(CODEC2_MODE_1300);
nsam = codec2_samples_per_frame(codec2);
buf = (short*)malloc(nsam*sizeof(short));
nbit = codec2_bits_per_frame(codec2);
bits = (unsigned char*)malloc(nbit*sizeof(char));
while(fread(buf, sizeof(short), nsam, fin) == (size_t)nsam) {
codec2_encode(codec2, bits, buf);
codec2_decode(codec2, buf, bits);
fwrite(buf, sizeof(short), nsam, fout);
}
free(buf);
free(bits);
codec2_destroy(codec2);
fclose(fin);
fclose(fout);
return 0;
}
static void c2demo(int mode, char inputfile[], char outputfile[])
{
struct CODEC2 *codec2;
short *inbuf, *outbuf;
unsigned char *bits;
int nsam, nbit;
FILE *fin, *fout;
int frame;
PROFILE_VAR(enc_start, dec_start);
codec2 = codec2_create(mode);
nsam = codec2_samples_per_frame(codec2);
outbuf = (short*)malloc(nsam*sizeof(short));
inbuf = (short*)malloc(nsam*sizeof(short));
nbit = codec2_bits_per_frame(codec2);
bits = (unsigned char*)malloc(nbit*sizeof(char));
fin = fopen(inputfile, "rb");
if (fin == NULL) {
printf("Error opening input file: %s\n\nTerminating....\n",inputfile);
exit(1);
}
fout = fopen(outputfile, "wb");
if (fout == NULL) {
printf("Error opening output file: %s\n\nTerminating....\n",outputfile);
exit(1);
}
#ifdef DUMP
dump_on("stm32f4");
#endif
frame = 0;
while (fread(inbuf, sizeof(short), nsam, fin) == nsam) {
PROFILE_SAMPLE(enc_start);
codec2_encode(codec2, bits, inbuf);
PROFILE_SAMPLE_AND_LOG(dec_start, enc_start, " enc");
codec2_decode(codec2, outbuf, bits);
PROFILE_SAMPLE_AND_LOG2(dec_start, " dec");
PROFILE_SAMPLE_AND_LOG2(enc_start, " enc & dec");
fwrite((char*)outbuf, sizeof(short), nsam, fout);
printf("frame: %d\n", ++frame);
machdep_profile_print_logged_samples();
}
#ifdef DUMP
dump_off("sm32f4");
#endif
fclose(fin);
fclose(fout);
free(inbuf);
free(outbuf);
free(bits);
codec2_destroy(codec2);
}
void AbsSeq::dump() { dump_on(gclog_or_tty); }
int main(int argc, char *argv[])
{
FILE *fout = NULL; /* output speech file */
FILE *fin; /* input speech file */
short buf[N]; /* input/output buffer */
float buf_float[N];
float Sn[M]; /* float input speech samples */
float Sn_pre[N]; /* pre-emphasised input speech samples */
COMP Sw[FFT_ENC]; /* DFT of Sn[] */
kiss_fft_cfg fft_fwd_cfg;
kiss_fft_cfg fft_inv_cfg;
float w[M]; /* time domain hamming window */
COMP W[FFT_ENC]; /* DFT of w[] */
MODEL model;
float Pn[2*N]; /* trapezoidal synthesis window */
float Sn_[2*N]; /* synthesised speech */
int i,m; /* loop variable */
int frames;
float prev_Wo, prev__Wo, prev_uq_Wo;
float pitch;
char out_file[MAX_STR];
char ampexp_arg[MAX_STR];
char phaseexp_arg[MAX_STR];
float snr;
float sum_snr;
int orderi;
int lpc_model = 0, order = LPC_ORD;
int lsp = 0, lspd = 0, lspvq = 0;
int lspres = 0;
int lspjvm = 0, lspjnd = 0, lspmel = 0;
#ifdef __EXPERIMENTAL__
int lspanssi = 0,
#endif
int prede = 0;
float pre_mem = 0.0, de_mem = 0.0;
float ak[order];
COMP Sw_[FFT_ENC];
COMP Ew[FFT_ENC];
int phase0 = 0;
float ex_phase[MAX_AMP+1];
int postfilt;
int hand_voicing = 0, phaseexp = 0, ampexp = 0, hi = 0, simlpcpf = 0;
int lpcpf = 0;
FILE *fvoicing = 0;
MODEL prev_model;
int dec;
int decimate = 1;
float lsps[order];
float e, prev_e;
int lsp_indexes[order];
float lsps_[order];
float Woe_[2];
float lsps_dec[4][LPC_ORD], e_dec[4], weight, weight_inc, ak_dec[4][LPC_ORD];
MODEL model_dec[4], prev_model_dec;
float prev_lsps_dec[order], prev_e_dec;
void *nlp_states;
float hpf_states[2];
int scalar_quant_Wo_e = 0;
int vector_quant_Wo_e = 0;
int dump_pitch_e = 0;
FILE *fjvm = NULL;
#ifdef DUMP
int dump;
#endif
struct PEXP *pexp = NULL;
struct AEXP *aexp = NULL;
float gain = 1.0;
int bpf_en = 0;
float bpf_buf[BPF_N+N];
char* opt_string = "ho:";
struct option long_options[] = {
{ "lpc", required_argument, &lpc_model, 1 },
{ "lspjnd", no_argument, &lspjnd, 1 },
{ "lspmel", no_argument, &lspmel, 1 },
{ "lsp", no_argument, &lsp, 1 },
{ "lspd", no_argument, &lspd, 1 },
{ "lspvq", no_argument, &lspvq, 1 },
{ "lspres", no_argument, &lspres, 1 },
{ "lspjvm", no_argument, &lspjvm, 1 },
#ifdef __EXPERIMENTAL__
{ "lspanssi", no_argument, &lspanssi, 1 },
#endif
{ "phase0", no_argument, &phase0, 1 },
{ "phaseexp", required_argument, &phaseexp, 1 },
{ "ampexp", required_argument, &exp, 1 },
{ "postfilter", no_argument, &postfilt, 1 },
{ "hand_voicing", required_argument, &hand_voicing, 1 },
{ "dec", required_argument, &dec, 1 },
{ "hi", no_argument, &hi, 1 },
{ "simlpcpf", no_argument, &simlpcpf, 1 },
{ "lpcpf", no_argument, &lpcpf, 1 },
{ "prede", no_argument, &prede, 1 },
{ "dump_pitch_e", required_argument, &dump_pitch_e, 1 },
{ "sq_pitch_e", no_argument, &scalar_quant_Wo_e, 1 },
{ "vq_pitch_e", no_argument, &vector_quant_Wo_e, 1 },
{ "rate", required_argument, NULL, 0 },
{ "gain", required_argument, NULL, 0 },
{ "bpf", no_argument, &bpf_en, 1 },
#ifdef DUMP
{ "dump", required_argument, &dump, 1 },
#endif
{ "help", no_argument, NULL, 'h' },
{ NULL, no_argument, NULL, 0 }
};
int num_opts=sizeof(long_options)/sizeof(struct option);
COMP Aw[FFT_ENC];
for(i=0; i<M; i++) {
Sn[i] = 1.0;
Sn_pre[i] = 1.0;
}
for(i=0; i<2*N; i++)
Sn_[i] = 0;
prev_uq_Wo = prev_Wo = prev__Wo = TWO_PI/P_MAX;
prev_model.Wo = TWO_PI/P_MIN;
prev_model.L = floor(PI/prev_model.Wo);
for(i=1; i<=prev_model.L; i++) {
prev_model.A[i] = 0.0;
prev_model.phi[i] = 0.0;
}
for(i=1; i<=MAX_AMP; i++) {
//ex_phase[i] = (PI/3)*(float)rand()/RAND_MAX;
ex_phase[i] = 0.0;
}
e = prev_e = 1;
hpf_states[0] = hpf_states[1] = 0.0;
nlp_states = nlp_create(M);
if (argc < 2) {
print_help(long_options, num_opts, argv);
}
/*----------------------------------------------------------------*\
Interpret Command Line Arguments
\*----------------------------------------------------------------*/
while(1) {
int option_index = 0;
int opt = getopt_long(argc, argv, opt_string,
long_options, &option_index);
if (opt == -1)
break;
switch (opt) {
case 0:
if(strcmp(long_options[option_index].name, "lpc") == 0) {
orderi = atoi(optarg);
if((orderi < 4) || (orderi > order)) {
fprintf(stderr, "Error in LPC order (4 to %d): %s\n", order, optarg);
exit(1);
}
order = orderi;
#ifdef DUMP
} else if(strcmp(long_options[option_index].name, "dump") == 0) {
if (dump)
dump_on(optarg);
#endif
} else if(strcmp(long_options[option_index].name, "lsp") == 0
|| strcmp(long_options[option_index].name, "lspd") == 0
|| strcmp(long_options[option_index].name, "lspvq") == 0) {
assert(order == LPC_ORD);
} else if(strcmp(long_options[option_index].name, "dec") == 0) {
decimate = atoi(optarg);
if ((decimate != 2) && (decimate != 4)) {
fprintf(stderr, "Error in --dec, must be 2 or 4\n");
exit(1);
}
if (!phase0) {
printf("needs --phase0 to resample phase when using --dec\n");
exit(1);
}
if (!lpc_model) {
printf("needs --lpc [order] to resample amplitudes when using --dec\n");
exit(1);
}
} else if(strcmp(long_options[option_index].name, "hand_voicing") == 0) {
if ((fvoicing = fopen(optarg,"rt")) == NULL) {
fprintf(stderr, "Error opening voicing file: %s: %s.\n",
optarg, strerror(errno));
exit(1);
}
} else if(strcmp(long_options[option_index].name, "dump_pitch_e") == 0) {
if ((fjvm = fopen(optarg,"wt")) == NULL) {
fprintf(stderr, "Error opening pitch & energy dump file: %s: %s.\n",
optarg, strerror(errno));
exit(1);
}
} else if(strcmp(long_options[option_index].name, "phaseexp") == 0) {
strcpy(phaseexp_arg, optarg);
} else if(strcmp(long_options[option_index].name, "ampexp") == 0) {
strcpy(ampexp_arg, optarg);
} else if(strcmp(long_options[option_index].name, "gain") == 0) {
gain = atof(optarg);
} else if(strcmp(long_options[option_index].name, "rate") == 0) {
if(strcmp(optarg,"3200") == 0) {
lpc_model = 1;
scalar_quant_Wo_e = 1;
lspd = 1;
phase0 = 1;
postfilt = 1;
decimate = 1;
lpcpf = 1;
} else if(strcmp(optarg,"2400") == 0) {
lpc_model = 1;
vector_quant_Wo_e = 1;
lsp = 1;
phase0 = 1;
postfilt = 1;
decimate = 2;
lpcpf = 1;
} else if(strcmp(optarg,"1400") == 0) {
lpc_model = 1;
vector_quant_Wo_e = 1;
lsp = 1;
phase0 = 1;
postfilt = 1;
decimate = 4;
lpcpf = 1;
} else if(strcmp(optarg,"1300") == 0) {
lpc_model = 1;
scalar_quant_Wo_e = 1;
lsp = 1;
phase0 = 1;
postfilt = 1;
decimate = 4;
lpcpf = 1;
} else if(strcmp(optarg,"1200") == 0) {
lpc_model = 1;
scalar_quant_Wo_e = 1;
lspjvm = 1;
phase0 = 1;
postfilt = 1;
decimate = 4;
lpcpf = 1;
} else {
fprintf(stderr, "Error: invalid output rate (3200|2400|1400|1200) %s\n", optarg);
exit(1);
}
}
break;
case 'h':
print_help(long_options, num_opts, argv);
break;
case 'o':
if (strcmp(optarg, "-") == 0) fout = stdout;
else if ((fout = fopen(optarg,"wb")) == NULL) {
fprintf(stderr, "Error opening output speech file: %s: %s.\n",
optarg, strerror(errno));
exit(1);
}
strcpy(out_file,optarg);
break;
default:
/* This will never be reached */
break;
}
}
/* Input file */
if (strcmp(argv[optind], "-") == 0) fin = stdin;
else if ((fin = fopen(argv[optind],"rb")) == NULL) {
fprintf(stderr, "Error opening input speech file: %s: %s.\n",
argv[optind], strerror(errno));
exit(1);
}
ex_phase[0] = 0;
Woe_[0] = Woe_[1] = 1.0;
/*
printf("lspd: %d lspdt: %d lspdt_mode: %d phase0: %d postfilt: %d "
"decimate: %d dt: %d\n",lspd,lspdt,lspdt_mode,phase0,postfilt,
decimate,dt);
*/
/* Initialise ------------------------------------------------------------*/
fft_fwd_cfg = kiss_fft_alloc(FFT_ENC, 0, NULL, NULL); /* fwd FFT,used in several places */
fft_inv_cfg = kiss_fft_alloc(FFT_DEC, 1, NULL, NULL); /* inverse FFT, used just for synth */
make_analysis_window(fft_fwd_cfg, w, W);
make_synthesis_window(Pn);
quantise_init();
if (phaseexp)
pexp = phase_experiment_create();
if (ampexp)
aexp = amp_experiment_create();
if (bpf_en) {
for(i=0; i<BPF_N; i++)
bpf_buf[i] = 0.0;
}
for(i=0; i<LPC_ORD; i++) {
prev_lsps_dec[i] = i*PI/(LPC_ORD+1);
}
prev_e_dec = 1;
for(m=1; m<=MAX_AMP; m++)
prev_model_dec.A[m] = 0.0;
prev_model_dec.Wo = TWO_PI/P_MAX;
prev_model_dec.L = PI/prev_model_dec.Wo;
prev_model_dec.voiced = 0;
/*----------------------------------------------------------------* \
Main Loop
\*----------------------------------------------------------------*/
frames = 0;
sum_snr = 0;
while(fread(buf,sizeof(short),N,fin)) {
frames++;
for(i=0; i<N; i++)
buf_float[i] = buf[i];
/* optionally filter input speech */
if (prede) {
pre_emp(Sn_pre, buf_float, &pre_mem, N);
for(i=0; i<N; i++)
buf_float[i] = Sn_pre[i];
}
if (bpf_en) {
for(i=0; i<BPF_N; i++)
bpf_buf[i] = bpf_buf[N+i];
for(i=0; i<N; i++)
bpf_buf[BPF_N+i] = buf_float[i];
inverse_filter(&bpf_buf[BPF_N], bpf, N, buf_float, BPF_N);
}
/* shift buffer of input samples, and insert new samples */
for(i=0; i<M-N; i++) {
Sn[i] = Sn[i+N];
}
for(i=0; i<N; i++)
Sn[i+M-N] = buf_float[i];
/*------------------------------------------------------------*\
Estimate Sinusoidal Model Parameters
\*------------------------------------------------------------*/
nlp(nlp_states,Sn,N,P_MIN,P_MAX,&pitch,Sw,W,&prev_uq_Wo);
model.Wo = TWO_PI/pitch;
dft_speech(fft_fwd_cfg, Sw, Sn, w);
two_stage_pitch_refinement(&model, Sw);
estimate_amplitudes(&model, Sw, W, 1);
#ifdef DUMP
dump_Sn(Sn); dump_Sw(Sw); dump_model(&model);
#endif
if (ampexp)
amp_experiment(aexp, &model, ampexp_arg);
if (phaseexp) {
#ifdef DUMP
dump_phase(&model.phi[0], model.L);
#endif
phase_experiment(pexp, &model, phaseexp_arg);
#ifdef DUMP
dump_phase_(&model.phi[0], model.L);
#endif
}
if (hi) {
int m;
for(m=1; m<model.L/2; m++)
model.A[m] = 0.0;
for(m=3*model.L/4; m<=model.L; m++)
model.A[m] = 0.0;
}
/*------------------------------------------------------------*\
Zero-phase modelling
\*------------------------------------------------------------*/
if (phase0) {
float Wn[M]; /* windowed speech samples */
float Rk[order+1]; /* autocorrelation coeffs */
COMP a[FFT_ENC];
#ifdef DUMP
dump_phase(&model.phi[0], model.L);
#endif
/* find aks here, these are overwritten if LPC modelling is enabled */
for(i=0; i<M; i++)
Wn[i] = Sn[i]*w[i];
autocorrelate(Wn,Rk,M,order);
levinson_durbin(Rk,ak,order);
/* determine voicing */
snr = est_voicing_mbe(&model, Sw, W, Sw_, Ew);
if (dump_pitch_e)
fprintf(fjvm, "%f %f %d ", model.Wo, snr, model.voiced);
//printf("snr %3.2f v: %d Wo: %f prev_Wo: %f\n", snr, model.voiced,
// model.Wo, prev_uq_Wo);
#ifdef DUMP
dump_Sw_(Sw_);
dump_Ew(Ew);
dump_snr(snr);
#endif
/* just to make sure we are not cheating - kill all phases */
for(i=0; i<=MAX_AMP; i++)
model.phi[i] = 0;
/* Determine DFT of A(exp(jw)), which is needed for phase0 model when
LPC is not used, e.g. indecimate=1 (10ms) frames with no LPC */
for(i=0; i<FFT_ENC; i++) {
a[i].real = 0.0;
a[i].imag = 0.0;
}
for(i=0; i<=order; i++)
a[i].real = ak[i];
kiss_fft(fft_fwd_cfg, (kiss_fft_cpx *)a, (kiss_fft_cpx *)Aw);
if (hand_voicing) {
fscanf(fvoicing,"%d\n",&model.voiced);
}
}
/*------------------------------------------------------------*\
LPC model amplitudes and LSP quantisation
\*------------------------------------------------------------*/
if (lpc_model) {
e = speech_to_uq_lsps(lsps, ak, Sn, w, order);
for(i=0; i<LPC_ORD; i++)
lsps_[i] = lsps[i];
#ifdef DUMP
dump_ak(ak, order);
dump_E(e);
#endif
/* tracking down -ve energy values with BW expansion */
/*
if (e < 0.0) {
int i;
FILE*f=fopen("x.txt","wt");
for(i=0; i<M; i++)
fprintf(f,"%f\n", Sn[i]);
fclose(f);
printf("e = %f frames = %d\n", e, frames);
for(i=0; i<order; i++)
printf("%f ", ak[i]);
exit(0);
}
*/
if (dump_pitch_e)
fprintf(fjvm, "%f\n", e);
#ifdef DUMP
dump_lsp(lsps);
#endif
/* various LSP quantisation schemes */
if (lsp) {
encode_lsps_scalar(lsp_indexes, lsps, LPC_ORD);
decode_lsps_scalar(lsps_, lsp_indexes, LPC_ORD);
bw_expand_lsps(lsps_, LPC_ORD, 50.0, 100.0);
lsp_to_lpc(lsps_, ak, LPC_ORD);
}
if (lspd) {
encode_lspds_scalar(lsp_indexes, lsps, LPC_ORD);
decode_lspds_scalar(lsps_, lsp_indexes, LPC_ORD);
lsp_to_lpc(lsps_, ak, LPC_ORD);
}
#ifdef __EXPERIMENTAL__
if (lspvq) {
lspvq_quantise(lsps, lsps_, LPC_ORD);
bw_expand_lsps(lsps_, LPC_ORD, 50.0, 100.0);
lsp_to_lpc(lsps_, ak, LPC_ORD);
}
#endif
if (lspjvm) {
/* Jean-Marc's multi-stage, split VQ */
lspjvm_quantise(lsps, lsps_, LPC_ORD);
{
float lsps_bw[LPC_ORD];
memcpy(lsps_bw, lsps_, sizeof(float)*LPC_ORD);
bw_expand_lsps(lsps_bw, LPC_ORD, 50.0, 100.0);
lsp_to_lpc(lsps_bw, ak, LPC_ORD);
}
}
#ifdef __EXPERIMENTAL__
if (lspanssi) {
/* multi-stage VQ from Anssi Ramo OH3GDD */
lspanssi_quantise(lsps, lsps_, LPC_ORD, 5);
bw_expand_lsps(lsps_, LPC_ORD, 50.0, 100.0);
lsp_to_lpc(lsps_, ak, LPC_ORD);
}
#endif
/* experimenting with non-linear LSP spacing to see if
it's just noticable */
if (lspjnd) {
for(i=0; i<LPC_ORD; i++)
lsps_[i] = lsps[i];
locate_lsps_jnd_steps(lsps_, LPC_ORD);
lsp_to_lpc(lsps_, ak, LPC_ORD);
}
/* Another experiment with non-linear LSP spacing, this
time using a scaled version of mel frequency axis
warping. The scaling is such that the integer output
can be directly sent over the channel.
*/
if (lspmel) {
float f, f_;
float mel[LPC_ORD];
int mel_indexes[LPC_ORD];
for(i=0; i<order; i++) {
f = (4000.0/PI)*lsps[i];
mel[i] = floor(2595.0*log10(1.0 + f/700.0) + 0.5);
}
for(i=1; i<order; i++) {
if (mel[i] == mel[i-1])
mel[i]++;
}
encode_mels_scalar(mel_indexes, mel, 6);
decode_mels_scalar(mel, mel_indexes, 6);
#ifdef DUMP
dump_mel(mel, order);
#endif
for(i=0; i<LPC_ORD; i++) {
f_ = 700.0*( pow(10.0, (float)mel[i]/2595.0) - 1.0);
lsps_[i] = f_*(PI/4000.0);
}
lsp_to_lpc(lsps_, ak, order);
}
if (scalar_quant_Wo_e) {
e = decode_energy(encode_energy(e, E_BITS), E_BITS);
model.Wo = decode_Wo(encode_Wo(model.Wo, WO_BITS), WO_BITS);
model.L = PI/model.Wo; /* if we quantise Wo re-compute L */
}
if (vector_quant_Wo_e) {
/* JVM's experimental joint Wo & LPC energy quantiser */
quantise_WoE(&model, &e, Woe_);
}
}
/*------------------------------------------------------------*\
Synthesise and optional decimation to 20 or 40ms frame rate
\*------------------------------------------------------------*/
/*
if decimate == 2, we interpolate frame n from frame n-1 and n+1
if decimate == 4, we interpolate frames n, n+1, n+2, from frames n-1 and n+3
This is meant to give identical results to the implementations of various modes
in codec2.c
*/
/* delay line to keep frame by frame voicing decisions */
for(i=0; i<decimate-1; i++)
model_dec[i] = model_dec[i+1];
model_dec[decimate-1] = model;
if ((frames % decimate) == 0) {
for(i=0; i<order; i++)
lsps_dec[decimate-1][i] = lsps_[i];
e_dec[decimate-1] = e;
model_dec[decimate-1] = model;
/* interpolate the model parameters */
weight_inc = 1.0/decimate;
for(i=0, weight=weight_inc; i<decimate-1; i++, weight += weight_inc) {
//model_dec[i].voiced = model_dec[decimate-1].voiced;
interpolate_lsp_ver2(&lsps_dec[i][0], prev_lsps_dec, &lsps_dec[decimate-1][0], weight, order);
interp_Wo2(&model_dec[i], &prev_model_dec, &model_dec[decimate-1], weight);
e_dec[i] = interp_energy2(prev_e_dec, e_dec[decimate-1],weight);
}
/* then recover spectral amplitudes and synthesise */
for(i=0; i<decimate; i++) {
if (lpc_model) {
lsp_to_lpc(&lsps_dec[i][0], &ak_dec[i][0], order);
aks_to_M2(fft_fwd_cfg, &ak_dec[i][0], order, &model_dec[i], e_dec[i],
&snr, 0, simlpcpf, lpcpf, 1, LPCPF_BETA, LPCPF_GAMMA, Aw);
apply_lpc_correction(&model_dec[i]);
#ifdef DUMP
dump_lsp_(&lsps_dec[i][0]);
dump_ak_(&ak_dec[i][0], order);
sum_snr += snr;
dump_quantised_model(&model_dec[i]);
#endif
}
if (phase0)
phase_synth_zero_order(fft_fwd_cfg, &model_dec[i], ex_phase, Aw);
synth_one_frame(fft_inv_cfg, buf, &model_dec[i], Sn_, Pn, prede, &de_mem, gain);
if (fout != NULL) fwrite(buf,sizeof(short),N,fout);
}
/*
for(i=0; i<decimate; i++) {
printf("%d Wo: %f L: %d v: %d\n", frames, model_dec[i].Wo, model_dec[i].L, model_dec[i].voiced);
}
if (frames == 4*50)
exit(0);
*/
/* update memories for next frame ----------------------------*/
prev_model_dec = model_dec[decimate-1];
prev_e_dec = e_dec[decimate-1];
for(i=0; i<LPC_ORD; i++)
prev_lsps_dec[i] = lsps_dec[decimate-1][i];
}
}
/*----------------------------------------------------------------*\
End Main Loop
\*----------------------------------------------------------------*/
fclose(fin);
if (fout != NULL)
fclose(fout);
if (lpc_model)
printf("SNR av = %5.2f dB\n", sum_snr/frames);
if (phaseexp)
phase_experiment_destroy(pexp);
if (ampexp)
amp_experiment_destroy(aexp);
#ifdef DUMP
if (dump)
dump_off();
#endif
if (hand_voicing)
fclose(fvoicing);
nlp_destroy(nlp_states);
return 0;
}
void dump() const {
dump_on(tty);
}
int main(int argc, char *argv[])
{
int mode;
void *codec2;
FILE *fin;
FILE *fout;
FILE *fber = NULL;
short *buf;
unsigned char *bits;
float *softdec_bits;
int nsam, nbit, nbyte, i, byte, frames, bits_proc, bit_errors, error_mode;
int nstart_bit, nend_bit, bit_rate;
int state, next_state;
float ber, r, burst_length, burst_period, burst_timer, ber_est;
unsigned char mask;
int natural, dump, softdec, bit, ret;
char* opt_string = "h:";
struct option long_options[] = {
{ "ber", required_argument, NULL, 0 },
{ "startbit", required_argument, NULL, 0 },
{ "endbit", required_argument, NULL, 0 },
{ "berfile", required_argument, NULL, 0 },
{ "natural", no_argument, &natural, 1 },
{ "softdec", no_argument, &softdec, 1 },
#ifdef DUMP
{ "dump", required_argument, &dump, 1 },
#endif
{ "help", no_argument, NULL, 'h' },
{ NULL, no_argument, NULL, 0 }
};
int num_opts=sizeof(long_options)/sizeof(struct option);
if (argc < 4)
print_help(long_options, num_opts, argv);
if (strcmp(argv[1],"3200") == 0)
mode = CODEC2_MODE_3200;
else if (strcmp(argv[1],"2400") == 0)
mode = CODEC2_MODE_2400;
else if (strcmp(argv[1],"1600") == 0)
mode = CODEC2_MODE_1600;
else if (strcmp(argv[1],"1400") == 0)
mode = CODEC2_MODE_1400;
else if (strcmp(argv[1],"1300") == 0)
mode = CODEC2_MODE_1300;
else if (strcmp(argv[1],"1200") == 0)
mode = CODEC2_MODE_1200;
else if (strcmp(argv[1],"700") == 0)
mode = CODEC2_MODE_700;
else {
fprintf(stderr, "Error in mode: %s. Must be 3200, 2400, 1600, 1400, 1300, 1200, or 700\n", argv[1]);
exit(1);
}
bit_rate = atoi(argv[1]);
if (strcmp(argv[2], "-") == 0) fin = stdin;
else if ( (fin = fopen(argv[2],"rb")) == NULL ) {
fprintf(stderr, "Error opening input bit file: %s: %s.\n",
argv[2], strerror(errno));
exit(1);
}
if (strcmp(argv[3], "-") == 0) fout = stdout;
else if ( (fout = fopen(argv[3],"wb")) == NULL ) {
fprintf(stderr, "Error opening output speech file: %s: %s.\n",
argv[3], strerror(errno));
exit(1);
}
error_mode = NONE;
ber = 0.0;
burst_length = burst_period = 0.0;
burst_timer = 0.0;
dump = natural = softdec = 0;
codec2 = codec2_create(mode);
nsam = codec2_samples_per_frame(codec2);
nbit = codec2_bits_per_frame(codec2);
buf = (short*)malloc(nsam*sizeof(short));
nbyte = (nbit + 7) / 8;
bits = (unsigned char*)malloc(nbyte*sizeof(char));
softdec_bits = (float*)malloc(nbit*sizeof(float));
frames = bit_errors = bits_proc = 0;
nstart_bit = 0;
nend_bit = nbit-1;
while(1) {
int option_index = 0;
int opt = getopt_long(argc, argv, opt_string,
long_options, &option_index);
if (opt == -1)
break;
switch (opt) {
case 0:
if(strcmp(long_options[option_index].name, "ber") == 0) {
ber = atof(optarg);
error_mode = UNIFORM;
} else if(strcmp(long_options[option_index].name, "startbit") == 0) {
nstart_bit = atoi(optarg);
} else if(strcmp(long_options[option_index].name, "endbit") == 0) {
nend_bit = atoi(optarg);
} else if(strcmp(long_options[option_index].name, "berfile") == 0) {
if ((fber = fopen(optarg,"wt")) == NULL) {
fprintf(stderr, "Error opening BER file: %s %s.\n",
optarg, strerror(errno));
exit(1);
}
}
#ifdef DUMP
else if(strcmp(long_options[option_index].name, "dump") == 0) {
if (dump)
dump_on(optarg);
}
#endif
break;
case 'h':
print_help(long_options, num_opts, argv);
break;
default:
/* This will never be reached */
break;
}
}
assert(nend_bit <= nbit);
codec2_set_natural_or_gray(codec2, !natural);
//printf("%d %d\n", nstart_bit, nend_bit);
//fprintf(stderr, "softdec: %d natural: %d\n", softdec, natural);
if (softdec)
ret = (fread(softdec_bits, sizeof(float), nbit, fin) == (size_t)nbit);
else
ret = (fread(bits, sizeof(char), nbyte, fin) == (size_t)nbyte);
while(ret) {
frames++;
// apply bit errors, MSB of byte 0 is bit 0 in frame, only works in packed mode
if ((error_mode == UNIFORM) || (error_mode == UNIFORM_RANGE)) {
assert(softdec == 0);
for(i=nstart_bit; i<nend_bit+1; i++) {
r = (float)rand()/RAND_MAX;
if (r < ber) {
byte = i/8;
//printf("nbyte %d nbit %d i %d byte %d bits[%d] 0x%0x ", nbyte, nbit, i, byte, byte, bits[byte]);
mask = 1 << (7 - i + byte*8);
bits[byte] ^= mask;
//printf("shift: %d mask: 0x%0x bits[%d] 0x%0x\n", 7 - i + byte*8, mask, byte, bits[byte] );
bit_errors++;
}
bits_proc++;
}
}
if (error_mode == TWO_STATE) {
assert(softdec == 0);
burst_timer += (float)nbit/bit_rate;
fprintf(stderr, "burst_timer: %f state: %d\n", burst_timer, state);
next_state = state;
switch(state) {
case 0:
/* clear channel state - no bit errors */
if (burst_timer > (burst_period - burst_length))
next_state = 1;
break;
case 1:
/* burst error state - 50% bit error rate */
for(i=nstart_bit; i<nend_bit+1; i++) {
r = (float)rand()/RAND_MAX;
if (r < 0.5) {
byte = i/8;
bits[byte] ^= 1 << (7 - i + byte*8);
bit_errors++;
}
bits_proc++;
}
if (burst_timer > burst_period) {
burst_timer = 0.0;
next_state = 0;
}
break;
}
state = next_state;
}
if (fber != NULL) {
if (fread(&ber_est, sizeof(float), 1, fber) != 1) {
fprintf(stderr, "ran out of BER estimates!\n");
exit(1);
}
//fprintf(stderr, "ber_est: %f\n", ber_est);
}
else
ber_est = 0.0;
if (softdec) {
/* pack bits, MSB received first */
bit = 7; byte = 0;
memset(bits, 0, nbyte);
for(i=0; i<nbit; i++) {
bits[byte] |= ((softdec_bits[i] < 0.0) << bit);
bit--;
if (bit < 0) {
bit = 7;
byte++;
}
}
codec2_set_softdec(codec2, softdec_bits);
}
codec2_decode_ber(codec2, buf, bits, ber_est);
fwrite(buf, sizeof(short), nsam, fout);
//if this is in a pipeline, we probably don't want the usual
//buffering to occur
if (fout == stdout) fflush(stdout);
if (fin == stdin) fflush(stdin);
if (softdec)
ret = (fread(softdec_bits, sizeof(float), nbit, fin) == (size_t)nbit);
else
ret = (fread(bits, sizeof(char), nbyte, fin) == (size_t)nbyte);
}
if (error_mode)
fprintf(stderr, "actual BER: %1.3f\n", (float)bit_errors/bits_proc);
codec2_destroy(codec2);
free(buf);
free(bits);
free(softdec_bits);
fclose(fin);
fclose(fout);
return 0;
}
