int main(int argc, char **argv)
{
cst_wave *nw, *all;
cst_val *files;
const cst_val *w;
cst_val *wavelist;
cst_features *args;
int i,j;
float ntime;
int stime;
const char *nwfile;
args = new_features();
files =
cst_args(argv,argc,
"usage: combine_waves OPTIONS\n"
"Combine waves into single waveform\n"
"-o <string> Output waveform\n"
"-f <int> Input sample rate (for raw input)\n"
"-itype <string> Input type, raw or headered\n"
"-wavelist <string> File containing times and wave filenames\n",
args);
wavelist = get_wavelist(get_param_string(args,"-wavelist","-"));
if (wavelist == 0)
return -1;
all = new_wave();
for (w = wavelist; w; w = val_cdr(w))
{
ntime = decode_time(val_string(val_car(w)));
nwfile = val_string(val_car(val_cdr(w)));
nw = new_wave();
if (cst_wave_load_riff(nw,nwfile) != CST_OK_FORMAT)
{
fprintf(stderr,
"combine_waves: can't read file or wrong format \"%s\"\n",
nwfile);
continue;
}
stime = ntime * nw->sample_rate;
cst_wave_resize(all,stime+nw->num_samples,1);
for (i=0,j=stime; i<nw->num_samples; i++,j++)
{
/* this will cause overflows */
all->samples[j] += nw->samples[i];
}
delete_wave(nw);
}
cst_wave_save_riff(all,get_param_string(args,"-o","-"));
return 0;
}
void test_create(void)
{
cst_wave *w = NULL;
w = new_wave();
TEST_CHECK(w != NULL);
TEST_CHECK(w->num_samples == 0);
TEST_CHECK(w->type == NULL);
delete_wave(w);
}
void test_resize(void)
{
cst_wave *w = new_wave();
cst_wave_resize(w, 200, 2);
TEST_CHECK(w->num_samples == 200);
TEST_CHECK(w->num_channels = 2);
TEST_CHECK(cst_wave_resize(NULL, 200, 2) < 0);
delete_wave(w);
}
void test_rescale(void)
{
int i;
cst_wave *w = new_wave();
cst_wave_resize(w, 10, 2);
for (i = 0; i < 10; i++)
w->samples[i] = 10 + i;
cst_wave_rescale(w, 65536 * 2); //scale x2
for (i = 0; i < 10; i++)
TEST_CHECK(w->samples[i] == (10 + i) * 2);
delete_wave(w);
}
cst_wave *lpc_resynth(cst_lpcres *lpcres)
{
cst_wave *w;
int i,j,r,o,k;
int ci,cr;
float *outbuf, *lpccoefs;
int pm_size_samps;
/* Get a new wave to build the signal into */
w = new_wave();
cst_wave_resize(w,lpcres->num_samples,1);
w->sample_rate = lpcres->sample_rate;
/* outbuf is a circular buffer with past relevant samples in it */
outbuf = cst_alloc(float,1+lpcres->num_channels);
/* unpacked lpc coefficients */
lpccoefs = cst_alloc(float,lpcres->num_channels);
for (r=0,o=lpcres->num_channels,i=0; i < lpcres->num_frames; i++)
{
pm_size_samps = lpcres->sizes[i];
/* Unpack the LPC coefficients */
for (k=0; k<lpcres->num_channels; k++)
{
lpccoefs[k] = (float)((((double)lpcres->frames[i][k])/65535.0)*
lpcres->lpc_range) + lpcres->lpc_min;
}
/* Note we don't zero the lead in from the previous part */
/* seems like you should but it makes it worse if you do */
/* memset(outbuf,0,sizeof(float)*(1+lpcres->num_channels)); */
/* resynthesis the signal */
for (j=0; j < pm_size_samps; j++,r++)
{
outbuf[o] = (float)cst_ulaw_to_short(lpcres->residual[r]);
cr = (o == 0 ? lpcres->num_channels : o-1);
for (ci=0; ci < lpcres->num_channels; ci++)
{
outbuf[o] += lpccoefs[ci] * outbuf[cr];
cr = (cr == 0 ? lpcres->num_channels : cr-1);
}
w->samples[r] = (short)(outbuf[o]);
o = (o == lpcres->num_channels ? 0 : o+1);
}
}
cst_free(outbuf);
cst_free(lpccoefs);
return w;
}
cst_wave *lpc_resynth_sfp(cst_lpcres *lpcres)
{
/* The fixed point spike excited, without floats */
cst_wave *w;
int i,j,r,o,k;
int ci,cr;
int *outbuf, *lpccoefs;
int pm_size_samps, ilpc_min, ilpc_range;
//int pp = 0;
/* Get a new wave to build the signal into */
w = new_wave();
cst_wave_resize(w,lpcres->num_samples,1);
w->sample_rate = lpcres->sample_rate;
/* outbuf is a circular buffer with past relevant samples in it */
outbuf = cst_alloc(int,1+lpcres->num_channels);
/* unpacked lpc coefficients */
lpccoefs = cst_alloc(int,lpcres->num_channels);
ilpc_min = (int)(lpcres->lpc_min*32768.0);
/* assume range is never > abs(16) */
ilpc_range = (int)(lpcres->lpc_range*2048.0);
for (r=0,o=lpcres->num_channels,i=0; i < lpcres->num_frames; i++)
{
pm_size_samps = lpcres->sizes[i];
/* Unpack the LPC coefficients */
for (k=0; k<lpcres->num_channels; k++)
lpccoefs[k]=((lpcres->frames[i][k]/2*ilpc_range)/2048+ilpc_min)/2;
/* resynthesis the signal */
for (j=0; j < pm_size_samps; j++,r++)
{
outbuf[o] = (int)cst_ulaw_to_short(lpcres->residual[r]);
cr = (o == 0 ? lpcres->num_channels : o-1);
for (ci=0; ci < lpcres->num_channels; ci++)
{
outbuf[o] += (lpccoefs[ci]*outbuf[cr])/16384;
cr = (cr == 0 ? lpcres->num_channels : cr-1);
}
w->samples[r] = (short)outbuf[o];
//pp = outbuf[o];
o = (o == lpcres->num_channels ? 0 : o+1);
}
}
cst_free(outbuf);
cst_free(lpccoefs);
return w;
}
float flite_ssml_to_speech(const char *filename,
cst_voice *voice,
const char *outtype)
{
cst_tokenstream *ts;
int fp;
cst_wave *w;
float d;
if ((ts = ts_open(filename,
get_param_string(voice->features,"text_whitespace",NULL),
get_param_string(voice->features,"text_singlecharsymbols",NULL),
get_param_string(voice->features,"text_prepunctuation",NULL),
get_param_string(voice->features,"text_postpunctuation",NULL)))
== NULL)
{
cst_errmsg("failed to open file \"%s\" for ssml reading\n",
filename);
return 1;
}
fp = get_param_int(voice->features,"file_start_position",0);
if (fp > 0)
ts_set_stream_pos(ts,fp);
/* If its a file to write to, create and save an empty wave file */
/* as we are going to incrementally append to it */
if (!cst_streq(outtype,"play") &&
!cst_streq(outtype,"none") &&
!cst_streq(outtype,"stream"))
{
w = new_wave();
cst_wave_resize(w,0,1);
cst_wave_set_sample_rate(w,16000);
cst_wave_save_riff(w,outtype); /* an empty wave */
delete_wave(w);
}
d = flite_ssml_to_speech_ts(ts,voice,outtype);
ts_close(ts);
return d;
}
cst_utterance *join_units_modified_lpc(cst_utterance *utt)
{
cst_wave *w = 0;
cst_lpcres *lpcres;
const char *resynth_type;
const cst_val *streaming_info_val;
resynth_type = get_param_string(utt->features, "resynth_type", "float");
f0_targets_to_pm(utt);
concat_units(utt);
lpcres = val_lpcres(utt_feat_val(utt, "target_lpcres"));
streaming_info_val = get_param_val(utt->features, "streaming_info", NULL);
if (streaming_info_val)
{
lpcres->asi = val_audio_streaming_info(streaming_info_val);
lpcres->asi->utt = utt;
}
if (cst_streq(resynth_type, "float"))
w = lpc_resynth(lpcres);
else if (cst_streq(resynth_type, "fixed"))
{
w = lpc_resynth_fixedpoint(lpcres);
}
else
{
cst_errmsg("unknown resynthesis type %s\n", resynth_type);
cst_error(); /* Should not happen */
}
if (w == NULL)
{
/* Synthesis Failed, probably because it was interrupted */
utt_set_feat_int(utt, "Interrupted", 1);
w = new_wave();
}
utt_set_wave(utt, w);
return utt;
}
int main(int argc, char **argv)
{
cst_wave *in, *out;
cst_val *files;
cst_features *args;
int i,j;
int w, a, t;
args = new_features();
files =
cst_args(argv,argc,
"usage: dcoffset_wave OPTIONS\n"
"Subtract window average from waveform\n"
"-i <string> Input waveform\n"
"-o <string> Output waveform\n"
"-w <int> Window size (in samples)\n",
args);
w = flite_get_param_int(args,"-w",20);
in = new_wave();
cst_wave_load_riff(in,flite_get_param_string(args,"-i","-"));
out = copy_wave(in);
for (i=0; i<=out->num_samples; i++)
{
for (t=a=0,j=i-w/2; j < i+w/2; j++)
{
if ((j > 0) && (j < out->num_samples))
{
t += 1;
a+=in->samples[j];
}
}
/* printf("%d %d %d %d %d\n",i,out->samples[i],a/t,t,out->samples[i]-a/t); */
out->samples[i] -= a/t;
}
cst_wave_save_riff(out,flite_get_param_string(args,"-o","-"));
return 0;
}
int main(int argc, char **argv)
{
cst_track *lpc;
cst_wave *sig, *sig2;
cst_sts *sts;
if (argc != 6)
{
fprintf(stderr,"usage: find_sts lpc_min lpc_range LPC WAVEFILE STS\n");
return 1;
}
lpc_min = atof(argv[1]);
lpc_range = atof(argv[2]);
lpc = new_track();
cst_track_load_est(lpc,argv[3]);
sig = new_wave();
if (cst_wave_load_riff(sig,argv[4]) == CST_WRONG_FORMAT)
{
fprintf(stderr,
"cannot load waveform, format unrecognized, from \"%s\"\n",
argv[4]);
exit(-1);
}
sts = find_sts(sig,lpc);
/* See if it worked */
sig2 = reconstruct_wave(sig,sts,lpc);
compare_waves(sig,sig2);
cst_wave_save_riff(sig2,"sig2.wav");
save_sts(sts,lpc,sig,argv[5]);
return 0;
}
int main(int argc, char **argv)
{
cst_wave *w;
cst_relation *r;
cst_utterance *u;
if (argc != 3)
{
fprintf(stderr, "usage: mimic_play_wave_sync WAVEFILE LABELFILE\n");
return 1;
}
w = new_wave();
if (cst_wave_load_riff(w, argv[1]) != CST_OK_FORMAT)
return -1;
u = new_utterance();
r = utt_relation_create(u, "FOO");
if (relation_load(r, argv[2]) != CST_OK_FORMAT)
return -1;
mimic_play_wave_sync(w, r, my_call_back);
return 0;
}
cst_wave *lpc_resynth_fixedpoint(cst_lpcres *lpcres)
{
/* The fixed point version, without floats */
cst_wave *w;
int i,j,r,o,k;
int stream_mark;
int ci,cr;
int *outbuf, *lpccoefs;
int pm_size_samps, ilpc_min, ilpc_range;
//int pp = 0;
int rc = CST_AUDIO_STREAM_CONT;
/* Get a new wave to build the signal into */
w = new_wave();
cst_wave_resize(w,lpcres->num_samples,1);
w->sample_rate = lpcres->sample_rate;
/* outbuf is a circular buffer with past relevant samples in it */
outbuf = cst_alloc(int,1+lpcres->num_channels);
/* unpacked lpc coefficients */
lpccoefs = cst_alloc(int,lpcres->num_channels);
ilpc_min = (int)(lpcres->lpc_min*32768.0);
/* assume range is never > abs(16) */
ilpc_range = (int)(lpcres->lpc_range*2048.0);
stream_mark = 0;
for (r=0,o=lpcres->num_channels,i=0;
(rc == CST_AUDIO_STREAM_CONT) && (i < lpcres->num_frames);
i++)
{
pm_size_samps = lpcres->sizes[i];
if (lpcres->delayed_decoding)
{
/* do decoding for this frame */
add_residual_g721vuv(lpcres->sizes[i],
&lpcres->residual[r],
lpcres->sizes[i],
lpcres->packed_residuals[i]);
}
/* Unpack the LPC coefficients */
for (k=0; k<lpcres->num_channels; k++)
lpccoefs[k]=((lpcres->frames[i][k]/2*ilpc_range)/2048+ilpc_min)/2;
/* resynthesis the signal */
for (j=0; j < pm_size_samps; j++,r++)
{
outbuf[o] = (int)ulaw_to_short_table[lpcres->residual[r]];
outbuf[o] *= 16384;
cr = (o == 0 ? lpcres->num_channels : o-1);
for (ci=0; ci < lpcres->num_channels; ci++)
{
outbuf[o] += lpccoefs[ci]*outbuf[cr];
cr = (cr == 0 ? lpcres->num_channels : cr-1);
}
outbuf[o] /= 16384;
w->samples[r] = (short)outbuf[o];
//pp = outbuf[o];
o = (o == lpcres->num_channels ? 0 : o+1);
}
if (lpcres->asi && (r-stream_mark > lpcres->asi->min_buffsize))
{
rc = (*lpcres->asi->asc)(w,stream_mark,r-stream_mark,0,
lpcres->asi);
stream_mark = r;
}
}
if ((lpcres->asi) && (rc == CST_AUDIO_STREAM_CONT))
(*lpcres->asi->asc)(w,stream_mark,r-stream_mark,1,lpcres->asi);
cst_free(outbuf);
cst_free(lpccoefs);
w->num_samples = r; /* just to be safe */
return w;
}
float flite_file_to_speech(const char *filename,
cst_voice *voice,
const char *outtype)
{
cst_utterance *utt;
cst_tokenstream *ts;
const char *token;
cst_item *t;
cst_relation *tokrel;
float d, durs = 0;
int num_tokens;
cst_breakfunc breakfunc = default_utt_break;
if ((ts = ts_open(filename,
get_param_string(voice->features,"text_whitespace",NULL),
get_param_string(voice->features,"text_singlecharsymbols",NULL),
get_param_string(voice->features,"text_prepunctuation",NULL),
get_param_string(voice->features,"text_postpunctuation",NULL)))
== NULL)
{
cst_errmsg("failed to open file \"%s\" for reading\n",
filename);
return 1;
}
if (feat_present(voice->features,"utt_break"))
breakfunc = val_breakfunc(feat_val(voice->features,"utt_break"));
/* If its a file to write to delete it as we're going to */
/* incrementally append to it */
if (!cst_streq(outtype,"play") && !cst_streq(outtype,"none"))
{
cst_wave *w;
w = new_wave();
cst_wave_resize(w,0,1);
cst_wave_set_sample_rate(w,16000);
cst_wave_save_riff(w,outtype); /* an empty wave */
delete_wave(w);
}
num_tokens = 0;
utt = new_utterance();
tokrel = utt_relation_create(utt, "Token");
while (!ts_eof(ts) || num_tokens > 0)
{
token = ts_get(ts);
if ((strlen(token) == 0) ||
(num_tokens > 500) || /* need an upper bound */
(relation_head(tokrel) &&
breakfunc(ts,token,tokrel)))
{
/* An end of utt */
d = flite_tokens_to_speech(utt,voice,outtype);
utt = NULL;
if (d < 0)
goto out;
durs += d;
if (ts_eof(ts))
goto out;
utt = new_utterance();
tokrel = utt_relation_create(utt, "Token");
num_tokens = 0;
}
num_tokens++;
t = relation_append(tokrel, NULL);
item_set_string(t,"name",token);
item_set_string(t,"whitespace",ts->whitespace);
item_set_string(t,"prepunctuation",ts->prepunctuation);
item_set_string(t,"punc",ts->postpunctuation);
item_set_int(t,"file_pos",ts->file_pos);
item_set_int(t,"line_number",ts->line_number);
}
out:
delete_utterance(utt);
ts_close(ts);
return durs;
}
int main(int argc, char **argv)
{
cst_wave *w;
int port;
char *server;
char *encoding;
int i,iw;
port = CST_AUDIO_DEFAULT_PORT;
server = CST_AUDIO_DEFAULT_SERVER;
encoding = CST_AUDIO_DEFAULT_ENCODING;
if (argc == 1)
{
play_client_main_usage();
return 1;
}
if ((cst_streq("-h",argv[1])) ||
(cst_streq("-help",argv[1])) ||
(cst_streq("--help",argv[1])))
{
play_client_main_usage();
return 1;
}
iw = 1;
if (cst_streq("-s",argv[iw]))
{
if (argc < iw+1)
{
fprintf(stderr,"ERROR: no servername given\n");
play_client_main_usage();
return 1;
}
server = argv[iw+1];
iw+=2;
}
if (cst_streq("-p",argv[iw]))
{
if (argc < iw+1)
{
fprintf(stderr,"ERROR: no port given\n");
play_client_main_usage();
return 1;
}
port = atoi(argv[iw+1]);
iw+=2;
}
if (cst_streq("-e",argv[iw]))
{
if (argc < iw+1)
{
fprintf(stderr,"ERROR: no encoding given\n");
play_client_main_usage();
return 1;
}
encoding = argv[iw+1];
iw+=2;
}
for (i=iw; i<argc; i++)
{
w = new_wave();
if (cst_wave_load_riff(w,argv[i]) != CST_OK_FORMAT)
{
fprintf(stderr,
"play_wave: can't read file or wrong format \"%s\"\n",
argv[i]);
continue;
}
play_wave_client(w,server,port,encoding);
delete_wave(w);
}
return 0;
}
float flite_file_to_speech(const char *filename,
cst_voice *voice,
const char *outtype)
{
cst_utterance *utt;
cst_tokenstream *ts;
const char *token;
cst_item *t;
cst_relation *tokrel;
float durs = 0;
int num_tokens;
cst_wave *w;
cst_breakfunc breakfunc = default_utt_break;
cst_uttfunc utt_user_callback = 0;
int fp;
if ((ts = ts_open(filename,
get_param_string(voice->features,"text_whitespace",NULL),
get_param_string(voice->features,"text_singlecharsymbols",NULL),
get_param_string(voice->features,"text_prepunctuation",NULL),
get_param_string(voice->features,"text_postpunctuation",NULL)))
== NULL)
{
cst_errmsg("failed to open file \"%s\" for reading\n",
filename);
return 1;
}
fp = get_param_int(voice->features,"file_start_position",0);
if (fp > 0)
ts_set_stream_pos(ts,fp);
if (feat_present(voice->features,"utt_break"))
breakfunc = val_breakfunc(feat_val(voice->features,"utt_break"));
if (feat_present(voice->features,"utt_user_callback"))
utt_user_callback = val_uttfunc(feat_val(voice->features,"utt_user_callback"));
/* If its a file to write to, create and save an empty wave file */
/* as we are going to incrementally append to it */
if (!cst_streq(outtype,"play") &&
!cst_streq(outtype,"none") &&
!cst_streq(outtype,"stream"))
{
w = new_wave();
cst_wave_resize(w,0,1);
cst_wave_set_sample_rate(w,16000);
cst_wave_save_riff(w,outtype); /* an empty wave */
delete_wave(w);
}
num_tokens = 0;
utt = new_utterance();
tokrel = utt_relation_create(utt, "Token");
while (!ts_eof(ts) || num_tokens > 0)
{
token = ts_get(ts);
if ((cst_strlen(token) == 0) ||
(num_tokens > 500) || /* need an upper bound */
(relation_head(tokrel) &&
breakfunc(ts,token,tokrel)))
{
/* An end of utt, so synthesize it */
if (utt_user_callback)
utt = (utt_user_callback)(utt);
if (utt)
{
utt = flite_do_synth(utt,voice,utt_synth_tokens);
durs += flite_process_output(utt,outtype,TRUE);
delete_utterance(utt); utt = NULL;
}
else
break;
if (ts_eof(ts)) break;
utt = new_utterance();
tokrel = utt_relation_create(utt, "Token");
num_tokens = 0;
}
num_tokens++;
t = relation_append(tokrel, NULL);
item_set_string(t,"name",token);
item_set_string(t,"whitespace",ts->whitespace);
item_set_string(t,"prepunctuation",ts->prepunctuation);
item_set_string(t,"punc",ts->postpunctuation);
/* Mark it at the beginning of the token */
item_set_int(t,"file_pos",
ts->file_pos-(1+ /* as we are already on the next char */
cst_strlen(token)+
cst_strlen(ts->prepunctuation)+
cst_strlen(ts->postpunctuation)));
item_set_int(t,"line_number",ts->line_number);
}
delete_utterance(utt);
ts_close(ts);
return durs;
}