/**
* Detect RIFF file and parse its header if detected.
*
* @return TRUE if it's a RIFF file, FALSE if not, -1 if an error occurred.
*/
static int
detect_riff(sphinx_wave2feat_t *wtf)
{
FILE *fh;
MSWAV_hdr hdr;
if ((fh = fopen(wtf->infile, "rb")) == NULL) {
E_ERROR_SYSTEM("Failed to open %s", wtf->infile);
return -1;
}
if (fread(&hdr, sizeof(hdr), 1, fh) != 1) {
E_ERROR_SYSTEM("Failed to read RIFF header");
fclose(fh);
return -1;
}
/* Make sure it is actually a RIFF file. */
if (0 != memcmp(hdr.rifftag, "RIFF", 4)) {
fclose(fh);
return FALSE;
}
/* Get relevant information. */
cmd_ln_set_int32_r(wtf->config, "-nchans", hdr.numchannels);
cmd_ln_set_float32_r(wtf->config, "-samprate", hdr.SamplingFreq);
wtf->infh = fh;
return TRUE;
}
static int
open_nist_file(sphinx_wave2feat_t *wtf, char const *infile, FILE **out_fh, int detect_endian)
{
char nist[7];
lineiter_t *li;
FILE *fh;
if ((fh = fopen(infile, "rb")) == NULL) {
E_ERROR_SYSTEM("Failed to open %s", infile);
return -1;
}
if (fread(&nist, 1, 7, fh) != 7) {
E_ERROR_SYSTEM("Failed to read NIST header");
fclose(fh);
return -1;
}
/* Is this actually a NIST file? */
if (0 != strncmp(nist, "NIST_1A", 7)) {
fclose(fh);
return FALSE;
}
/* Rewind, parse lines. */
fseek(fh, 0, SEEK_SET);
for (li = lineiter_start(fh); li; li = lineiter_next(li)) {
char **words;
int nword;
string_trim(li->buf, STRING_BOTH);
if (strlen(li->buf) == 0) {
lineiter_free(li);
break;
}
nword = str2words(li->buf, NULL, 0);
if (nword != 3)
continue;
words = (char **)ckd_calloc(nword, sizeof(*words));
str2words(li->buf, words, nword);
if (0 == strcmp(words[0], "sample_rate")) {
cmd_ln_set_float32_r(wtf->config, "-samprate", atof_c(words[2]));
}
if (0 == strcmp(words[0], "channel_count")) {
cmd_ln_set_int32_r(wtf->config, "-nchans", atoi(words[2]));
}
if (detect_endian && 0 == strcmp(words[0], "sample_byte_format")) {
cmd_ln_set_str_r(wtf->config, "-input_endian",
(0 == strcmp(words[2], "10")) ? "big" : "little");
}
ckd_free(words);
}
fseek(fh, 1024, SEEK_SET);
if (out_fh)
*out_fh = fh;
else
fclose(fh);
return TRUE;
}
/**
* Detect a file supported by libsndfile and parse its header if detected.
*
* @return TRUE if it's a supported file, FALSE if not, -1 if an error occurred.
*/
static int
detect_sndfile(sphinx_wave2feat_t *wtf)
{
SNDFILE *sf;
SF_INFO sfinfo;
memset(&sfinfo, 0, sizeof(sfinfo));
/* We let other detectors catch I/O errors, since there is
no way to tell them from format errors when opening :( */
if ((sf = sf_open(wtf->infile, SFM_READ, &sfinfo)) == NULL) {
return FALSE;
}
/* Get relevant information. */
cmd_ln_set_int32_r(wtf->config, "-nchans", sfinfo.channels);
cmd_ln_set_float32_r(wtf->config, "-samprate", sfinfo.samplerate);
wtf->insfh = sf;
wtf->infh = NULL;
return TRUE;
}
static void
gst_pocketsphinx_set_int(GstPocketSphinx *ps,
const gchar *key, const GValue *value)
{
cmd_ln_set_int32_r(ps->config, key, g_value_get_int(value));
}
int
sphinx_wave2feat_convert_file(sphinx_wave2feat_t *wtf,
char const *infile, char const *outfile)
{
int nchans, minfft, nfft, nfloat, veclen;
audio_type_t const *atype;
int fshift, fsize;
if (cmd_ln_boolean_r(wtf->config, "-verbose"))
E_INFO("Converting %s to %s\n", infile, outfile);
wtf->infile = ckd_salloc(infile);
/* Detect input file type. */
if ((atype = detect_audio_type(wtf)) == NULL)
return -1;
/* Determine whether to byteswap input. */
wtf->byteswap = strcmp(cmd_ln_str_r(wtf->config, "-mach_endian"),
cmd_ln_str_r(wtf->config, "-input_endian"));
/* Make sure the FFT size is sufficiently large. */
minfft = (int)(cmd_ln_float32_r(wtf->config, "-samprate")
* cmd_ln_float32_r(wtf->config, "-wlen") + 0.5);
for (nfft = 1; nfft < minfft; nfft <<= 1)
;
if (nfft > cmd_ln_int32_r(wtf->config, "-nfft")) {
E_WARN("Value of -nfft = %d is too small, increasing to %d\n",
cmd_ln_int32_r(wtf->config, "-nfft"), nfft);
cmd_ln_set_int32_r(wtf->config, "-nfft", nfft);
fe_free(wtf->fe);
wtf->fe = fe_init_auto_r(wtf->config);
}
/* Get the output frame size (if not already set). */
if (wtf->veclen == 0)
wtf->veclen = fe_get_output_size(wtf->fe);
/* Set up the input and output buffers. */
fe_get_input_size(wtf->fe, &fshift, &fsize);
/* Want to get at least a whole frame plus shift in here. Also we
will either pick or mix multiple channels so we need to read
them all at once. */
nchans = cmd_ln_int32_r(wtf->config, "-nchans");
wtf->blocksize = cmd_ln_int32_r(wtf->config, "-blocksize") * nchans;
if (wtf->blocksize < (fsize + fshift) * nchans) {
E_INFO("Block size of %d too small, increasing to %d\n",
wtf->blocksize,
(fsize + fshift) * nchans);
wtf->blocksize = (fsize + fshift) * nchans;
}
wtf->audio = (short *)ckd_calloc(wtf->blocksize, sizeof(*wtf->audio));
wtf->featsize = (wtf->blocksize / nchans - fsize) / fshift;
/* Use the maximum of the input and output frame sizes to allocate this. */
veclen = wtf->veclen;
if (wtf->in_veclen > veclen) veclen = wtf->in_veclen;
wtf->feat = (mfcc_t**)ckd_calloc_2d(wtf->featsize, veclen, sizeof(**wtf->feat));
/* Let's go! */
if ((wtf->outfh = fopen(outfile, "wb")) == NULL) {
E_ERROR_SYSTEM("Failed to open %s for writing", outfile);
return -1;
}
/* Write an empty header, which we'll fill in later. */
if (wtf->ot->output_header &&
(*wtf->ot->output_header)(wtf, 0) < 0) {
E_ERROR_SYSTEM("Failed to write empty header to %s\n", outfile);
goto error_out;
}
wtf->outfile = ckd_salloc(outfile);
if ((nfloat = (*atype->decode)(wtf)) < 0) {
E_ERROR("Failed to convert");
goto error_out;
}
if (wtf->ot->output_header) {
if (fseek(wtf->outfh, 0, SEEK_SET) < 0) {
E_ERROR_SYSTEM("Failed to seek to beginning of %s\n", outfile);
goto error_out;
}
if ((*wtf->ot->output_header)(wtf, nfloat) < 0) {
E_ERROR_SYSTEM("Failed to write header to %s\n", outfile);
goto error_out;
}
}
if (wtf->audio)
ckd_free(wtf->audio);
if (wtf->feat)
ckd_free_2d(wtf->feat);
if (wtf->infile)
ckd_free(wtf->infile);
if (wtf->outfile)
ckd_free(wtf->outfile);
wtf->audio = NULL;
wtf->infile = NULL;
wtf->feat = NULL;
wtf->outfile = NULL;
if (wtf->outfh)
if (fclose(wtf->outfh) == EOF)
E_ERROR_SYSTEM("Failed to close output file");
wtf->outfh = NULL;
return 0;
error_out:
if (wtf->audio)
ckd_free(wtf->audio);
if (wtf->feat)
ckd_free_2d(wtf->feat);
if (wtf->infile)
ckd_free(wtf->infile);
if (wtf->outfile)
ckd_free(wtf->outfile);
wtf->audio = NULL;
wtf->infile = NULL;
wtf->feat = NULL;
wtf->outfile = NULL;
if (wtf->outfh)
if (fclose(wtf->outfh) == EOF)
E_ERROR_SYSTEM("Failed to close output file");
wtf->outfh = NULL;
return -1;
}
