def export_dictionary(ts_all, lex, dictfn2, prompt_words):
logging.info("Exporting dictionary...")
utt_dict = {}
if prompt_words:
for ts in ts_all:
tsd = ts_all[ts]
tokens = tsd['ts'].split(' ')
# logging.info ( '%s %s' % (repr(ts), repr(tokens)) )
for token in tokens:
if token in utt_dict:
continue
if not token in lex:
logging.error(
"*** ERROR: missing token in dictionary: '%s' (tsd=%s, tokens=%s)"
% (token, repr(tsd), repr(tokens)))
sys.exit(1)
utt_dict[token] = lex.get_multi(token)
else:
for token in lex:
utt_dict[token] = lex.get_multi(token)
ps = {}
with open(dictfn2, 'w') as dictf:
dictf.write('!SIL SIL\n')
for token in sorted(utt_dict):
for form in utt_dict[token]:
ipa = utt_dict[token][form]['ipa']
xsr = ipa2xsampa(token, ipa, spaces=True)
xs = (xsr.replace('-', '').replace('\' ', '\'').replace(
' ', ' ').replace('#', 'nC'))
dictf.write((u'%s %s\n' % (token, xs)).encode('utf8'))
for p in xs.split(' '):
if len(p) < 1:
logging.error(
u"****ERROR: empty phoneme in : '%s' ('%s', ipa: '%s', token: '%s')"
% (xs, xsr, ipa, token))
pws = p[1:] if p[0] == '\'' else p
if not pws in ps:
ps[pws] = {p}
else:
ps[pws].add(p)
logging.info("%s written." % dictfn2)
logging.info("Exporting dictionary ... done.")
return ps, utt_dict
def test_ipa(self):
res = ipa2xsampa("EISENBAHN", u"ˈaɪ̯zən̩ˌbaːn")
#print "res: %s" % res
self.assertEqual(res, "'aIz@nba:n")
res = ipa2xsampa("DIPHTONGTEST", u"aɪɔɪaʊɜ'")
#print "res: %s" % res
self.assertEqual(res, "aIOIaU3")
res = ipa2mary("EISENBAHN", u"ˈaɪ̯zən̩ˌbaːn")
#print "res: %s" % res
self.assertEqual(res, "'aIz@nba:n")
res = ipa2mary("DIPHTONGTEST", u"aɪɔɪaʊɜ'")
#print "res: %s" % res
self.assertEqual(res, "aIOIaUr='")
def test_ipa(self):
res = ipa2xsampa("EISENBAHN", u"ˈaɪ̯zən̩ˌbaːn")
#print "res: %s" % res
self.assertEqual(res, "'aIz@nba:n")
res = ipa2xsampa("DIPHTONGTEST", u"aɪɔɪaʊɜ'")
#print "res: %s" % res
self.assertEqual(res, "aIOIaU3")
res = ipa2xsampa("BON", u"bɔ̃")
#print "res: %s" % res
self.assertEqual(res, "bO~")
res = ipa2xsampa("RESTAURANT", u"ʁɛstɔʁɑ̃")
#print "res: %s" % res
self.assertEqual(res, "REstORA~")
res = ipa2xsampa("VIN", u"vɛ̃")
#print "res: %s" % res
self.assertEqual(res, "vE~")
res = ipa2xsampa("BRUN", u"bʁœ̃")
#print "res: %s" % res
self.assertEqual(res, "bR9~")
res = ipa2xsampa("POIGNANT", u"pwaɲɑ̃")
#print "res: %s" % res
self.assertEqual(res, "pwaJA~")
res = ipa2mary("EISENBAHN", u"ˈaɪ̯zən̩ˌbaːn")
#print "res: %s" % res
self.assertEqual(res, "'aIz@nba:n")
res = ipa2mary("DIPHTONGTEST", u"aɪɔɪaʊɜ'")
#print "res: %s" % res
self.assertEqual(res, "aIOIaUr='")
res = ipa2mary("BON", u"bɔ̃")
#print "res: %s" % res
self.assertEqual(res, "bO~")
res = ipa2mary("RESTAURANT", u"ʁɛstɔʁɑ̃")
#print "res: %s" % res
self.assertEqual(res, "REstORA~")
res = ipa2mary("VIN", u"vɛ̃")
#print "res: %s" % res
self.assertEqual(res, "vE~")
res = ipa2mary("BRUN", u"bʁœ̃")
#print "res: %s" % res
self.assertEqual(res, "bR9~")
res = ipa2mary("POIGNANT", u"pwaɲɑ̃")
#print "res: %s" % res
self.assertEqual(res, "pwaJA~")
def gen_ipa(self, word):
if self._host_tts == 'local':
if self.engine == 'mary':
self.marytts.voice = self._voice
self.marytts.locale = self._locale
mp = self.marytts.g2p(word)
return mary2ipa(word, mp)
elif self.engine == 'espeak':
self.espeak.voice = self._voice
e_ipa = self.espeak.g2p(word, ipa='2')
xs = ipa2xsampa(word, e_ipa)
ipa = xsampa2ipa(word, xs)
logging.debug(u'espeak g2p: %s -> %s -> %s -> %s' %
(word, e_ipa, xs, ipa))
return ipa
elif self.engine == 'sequitur':
if not self.voice in SEQUITUR_MODELS:
raise Exception("no sequitur model for voice '%s'" %
self.voice)
return sequitur_gen_ipa(SEQUITUR_MODELS[self.voice], word)
else:
raise Exception("unknown engine '%s'" % self.engine)
else:
args = {
'l': self._locale,
'v': self._voice,
'e': self._engine,
't': word.encode('utf8')
}
url = 'http://%s:%s/tts/g2p?%s' % (self._host_tts, self._port_tts,
urllib.urlencode(args))
response = requests.get(url)
if response.status_code != 200:
return None
return response.json()['ipa']
def export_audio (train_val, tsdict):
global data_dir, utt_num, options
destdirfn = '%s/%s' % (data_dir, train_val)
for utt_id in tsdict:
with codecs.open('%s/%09d.id' % (destdirfn, utt_num[train_val]), 'w', 'utf8') as idf, \
codecs.open('%s/%09d.tkn' % (destdirfn, utt_num[train_val]), 'w', 'utf8') as tknf, \
codecs.open('%s/%09d.wrd' % (destdirfn, utt_num[train_val]), 'w', 'utf8') as wrdf :
ts = tsdict[utt_id]
tkn = u''
wrd = u''
for token in tokenize(ts['ts'], lang=options.lang):
if not (token in lex):
logging.error(u'token %s missing from dict!' % token)
logging.error(u'utt_id: %s' % utt_id)
logging.error(u'ts: %s' % ts['ts'])
sys.exit(1)
ipas = lex[token]['ipa']
xsr = ipa2xsampa(token, ipas, spaces=True)
xs = (xsr.replace('-', '')
.replace('\' ', '\'')
.replace(' ', ' ')
.replace('#', 'nC'))
if tkn:
tkn += u' | '
wrd += u' '
tkn += xs
wrd += token
tknf.write('%s\n' % tkn)
wrdf.write('%s\n' % wrd)
cmd = 'ln -s %s/%s/%s.wav %s/%09d.wav' % (wav16_dir, ts['corpus_name'], utt_id, destdirfn, utt_num[train_val])
logging.debug(cmd)
os.system(cmd)
# utt2spkf.write('%s %s\n' % (utt_id, ts['spk']))
utt_num[train_val] = utt_num[train_val] + 1
def lex_edit(lex_token):
global lex, lang
lex_base = lex_token.split('_')[0]
if lex_token in lex:
lex_entry = lex[lex_token]
else:
ipas = lex_gen_ipa(lex_base, 'de', 'sequitur', 'de')
lex_entry = {'ipa': ipas}
lex[lex_token] = lex_entry
ipas = lex_entry['ipa']
try:
tts.locale = 'de'
tts.engine = 'mary'
tts.voice = 'dfki-pavoque-neutral-hsmm'
tts.say_ipa(ipas, async=True)
except:
logging.error('EXCEPTION CAUGHT %s' % traceback.format_exc())
lex_gen = {}
lex_gen['de-mary'] = lex_gen_ipa(lex_base, 'de', 'mary', 'bits3')
lex_gen['de-espeak'] = lex_gen_ipa(lex_base, 'de', 'espeak', 'de')
lex_gen['de-sequitur'] = lex_gen_ipa(lex_base, 'de', 'sequitur', 'de')
while True:
print
print u"Token : %s" % lex_token
print u"IPA : %s" % lex_entry['ipa']
print
for engine in sorted(lex_gen):
print u"%-11s : %s" % (engine, lex_gen[engine])
print
if lex_token in lex:
m = lex.get_multi(lex_token)
for k in m:
print u"%s [%s]" % (k, m[k]['ipa'])
else:
print u"NEW TOKEN"
print u"SPEAK P:de-unitsel O:de-hsmm I:fr-hsmm U:en-hsmm"
print u"GEN G:de-mary H:de-espeak J:de-sequitur K:fr-mary L:en-mary"
print u" E:Edit Q:Quit "
try:
resp = raw_input("Lex> ")
# quit
if resp.lower() == 'q':
break
# generate de-mary
elif resp.lower() == 'g':
lex_entry['ipa'] = lex_gen_ipa(lex_base, 'de', 'mary', 'bits3',
True)
# generate de-espeak
elif resp.lower() == 'h':
lex_entry['ipa'] = lex_gen_ipa(lex_base, 'de', 'espeak', 'de',
True)
# generate en-mary
elif resp.lower() == 'l':
tts.locale = 'en-US'
tts.engine = 'mary'
tts.voice = 'cmu-rms-hsmm'
ipas = tts.gen_ipa(lex_base)
tts.say_ipa(ipas, async=True)
lex_entry['ipa'] = ipas
# generate fr-mary
elif resp.lower() == 'k':
tts.locale = 'fr'
tts.engine = 'mary'
tts.voice = 'upmc-pierre-hsmm'
ipas = tts.gen_ipa(lex_base)
tts.say_ipa(ipas, async=True)
lex_entry['ipa'] = ipas
# generate de-sequitur
elif resp.lower() == 'j':
lex_entry['ipa'] = lex_gen_ipa(lex_base, 'de', 'sequitur',
'de', True)
# speak de mary unitsel
elif resp.lower() == 'p':
if len(lex_entry['ipa']) == 0:
continue
ipas = lex_entry['ipa']
tts.locale = 'de'
tts.engine = 'mary'
tts.voice = 'bits3'
tts.say_ipa(ipas, async=True)
# speak de mary hsmm
elif resp.lower() == 'o':
if len(lex_entry['ipa']) == 0:
continue
ipas = lex_entry['ipa']
tts.locale = 'de'
tts.engine = 'mary'
tts.voice = 'dfki-pavoque-neutral-hsmm'
tts.say_ipa(ipas, async=True)
# speak fr mary hsmm
elif resp.lower() == 'i':
if len(lex_entry['ipa']) == 0:
continue
ipas = lex_entry['ipa']
tts.locale = 'fr'
tts.engine = 'mary'
tts.voice = 'upmc-pierre-hsmm'
tts.say_ipa(ipas, async=True)
# speak en mary hsmm
elif resp.lower() == 'u':
ipas = lex_entry['ipa']
tts.locale = 'en-US'
tts.engine = 'mary'
tts.voice = 'cmu-rms-hsmm'
tts.say_ipa(ipas, async=True)
# edit XS
elif resp.lower() == 'e':
ipas = lex_entry['ipa']
xs = ipa2xsampa(lex_token, ipas, stress_to_vowels=False)
readline.add_history(xs)
xs = raw_input(xs + '> ')
ipas = xsampa2ipa(lex_token, xs)
lex_entry['ipa'] = ipas
except:
logging.error('EXCEPTION CAUGHT %s' % traceback.format_exc())
lex.save()
print "new lexicon saved."
print
with codecs.open('%s/train.lex' % WORKDIR, 'w', 'utf8') as trainf, \
codecs.open('%s/test.lex' % WORKDIR, 'w', 'utf8') as testf, \
codecs.open('%s/all.lex' % WORKDIR, 'w', 'utf8') as allf :
cnt = 0
for token in lex:
if not token in wiktionary:
# print u"Missing in wiktionary: %s" % token
num_missing += 1
else:
num_found += 1
source_ipa = wiktionary[token][1]
target_ipa = lex[token]['ipa'].replace(u'-', u'')
target_xs = ipa2xsampa (token, target_ipa, spaces=True, stress_to_vowels=False)
if cnt % 10 == 0:
testf.write (u'%s %s\n' % (source_ipa, target_xs))
else:
trainf.write (u'%s %s\n' % (source_ipa, target_xs))
allf.write (u'%s %s\n' % (source_ipa, target_xs))
cnt += 1
logging.info('sequitur workdir %s done. %d entries.' % (WORKDIR, cnt))
lex = Lexicon(file_name=dict_name)
logging.info("loading lexicon...done.")
#
# export
#
misc.mkdirs(workdir)
with codecs.open('%s/train.lex' % workdir, 'w', 'utf8') as trainf, \
codecs.open('%s/test.lex' % workdir, 'w', 'utf8') as testf, \
codecs.open('%s/all.lex' % workdir, 'w', 'utf8') as allf :
cnt = 0
for word in lex:
ipa = lex[word]['ipa']
xs = ipa2xsampa(word, ipa, spaces=True, stress_to_vowels=False)
if cnt % 10 == 0:
testf.write(u'%s %s\n' % (word, xs))
else:
trainf.write(u'%s %s\n' % (word, xs))
allf.write(u'%s %s\n' % (word, xs))
cnt += 1
logging.info('sequitur workdir %s done.' % workdir)
def export_sphinx_case(work_dir, sphinxtrain_cfg_fn):
#
# language model
#
misc.mkdirs('%s' % work_dir)
fn = '%s/prompts.sent' % work_dir
with codecs.open(fn, 'w', 'utf8') as outf:
for cfn in ts_all:
transcript = transcripts[cfn]['ts']
outf.write('%s\n' % transcript)
logging.info("%s written." % fn)
fn = '%s/wlist.txt' % work_dir
with codecs.open(fn, 'w', 'utf8') as outf:
for word in lex:
if ENABLE_NOISE_FILLER:
if word == NOISE_WORD:
logging.debug('skipping noise word')
continue
outf.write('%s\n' % word)
logging.info("%s written." % fn)
#
# create work_dir structure
#
mfcc_dir = "%s/mfcc" % work_dir
misc.mkdirs('%s/logs' % work_dir)
misc.mkdirs('%s/etc' % work_dir)
misc.mkdirs('%s' % mfcc_dir)
# generate sphinx_train.cfg, featdir in there
# inf = codecs.open ('data/src/speech/sphinx_train.cfg', 'r', 'utf8')
# outf = codecs.open ('%s/etc/sphinx_train.cfg' % work_dir, 'w', 'utf8')
# for line in inf:
# s = line.replace('%FEATDIR%', mfcc_dir).replace('%WORKDIR%', work_dir)
# outf.write (s)
# inf.close()
# outf.close()
misc.copy_file(sphinxtrain_cfg_fn, '%s/etc/sphinx_train.cfg' % work_dir)
if ENABLE_NOISE_FILLER:
misc.copy_file('data/src/speech/sphinx-voxforge-noise.filler',
'%s/etc/voxforge.filler' % work_dir)
else:
misc.copy_file('data/src/speech/sphinx-voxforge.filler',
'%s/etc/voxforge.filler' % work_dir)
misc.copy_file('data/src/speech/sphinx-feat.params',
'%s/etc/feat.params' % work_dir)
#
# prompts
#
train_fifn = '%s/etc/voxforge_train.fileids' % work_dir
train_tsfn = '%s/etc/voxforge_train.transcription' % work_dir
test_fifn = '%s/etc/voxforge_test.fileids' % work_dir
test_tsfn = '%s/etc/voxforge_test.transcription' % work_dir
runfeatfn = '%s/run-feat.sh' % work_dir
lex_covered = set()
SPHINXFE = "sphinx_fe -i '%s' -part 1 -npart 1 -ei wav -o '%s' -eo mfc -nist no -raw no -mswav yes -samprate 16000 -lowerf 130 -upperf 6800 -nfilt 25 -transform dct -lifter 22 >>logs/mfcc%02d.log 2>&1 &\n"
with codecs.open(runfeatfn, 'w', 'utf8') as runfeatf:
runfeatf.write('#!/bin/bash\n\n')
cnt = 0
for cfn in ts_all:
w16filename = "%s/%s/%s.wav" % (wav16_dir, cfn2corpus[cfn], cfn)
mfcfilename = "mfcc/%s.mfc" % cfn
runfeatf.write(SPHINXFE % (w16filename, mfcfilename, cnt))
cnt = (cnt + 1) % NJOBS
if cnt == 0:
runfeatf.write('wait\n')
logging.info("%s written." % runfeatfn)
with codecs.open (train_fifn, 'w', 'utf8') as train_fif, \
codecs.open (train_tsfn, 'w', 'utf8') as train_tsf, \
codecs.open (test_fifn, 'w', 'utf8') as test_fif, \
codecs.open (test_tsfn, 'w', 'utf8') as test_tsf:
for cfn in ts_train:
train_fif.write('%s\n' % cfn)
tokens = tokenize(ts_train[cfn]['ts'],
lang=options.lang,
keep_punctuation=False)
ts = u' '.join(tokens)
train_tsf.write(u'<s> %s </s> (%s)\n' % (ts, cfn))
for token in tokens:
if not token in lex:
logging.error('word %s not covered by dict!')
sys.exit(1)
lex_covered.add(token)
for cfn in ts_test:
test_fif.write('%s\n' % cfn)
tokens = tokenize(ts_test[cfn]['ts'],
lang=options.lang,
keep_punctuation=False)
ts = u' '.join(tokens)
test_tsf.write(u'<s> %s </s> (%s)\n' % (ts, cfn))
for token in tokens:
if not token in lex:
logging.error('word %s not covered by dict!')
sys.exit(1)
lex_covered.add(token)
logging.info("%s written." % train_tsfn)
logging.info("%s written." % train_fifn)
logging.info("%s written." % test_tsfn)
logging.info("%s written." % test_fifn)
# generate dict
phoneset = set()
pdfn = '%s/etc/voxforge.dic' % work_dir
with codecs.open(pdfn, 'w', 'utf8') as pdf:
for word in lex:
if ENABLE_NOISE_FILLER:
if word == NOISE_WORD:
logging.debug('skipping noise word')
continue
if not word in lex_covered:
logging.debug(
'skipping word %s as it is not covered by transcripts' %
word)
continue
ipa = lex[word]['ipa']
xs = ipa2xsampa(word, ipa)
xa = xsampa2xarpabet(word, xs)
pdf.write(u'%s %s\n' % (word, xa))
phones = xa.split(' ')
for phone in phones:
if len(phone.strip()) == 0:
logging.error(
u"***ERROR: empty phone detected in lex entry %s %s" %
(word, ipa))
phoneset.add(phone)
logging.info("%s written." % pdfn)
logging.info("Got %d phones." % len(phoneset))
phfn = '%s/etc/voxforge.phone' % work_dir
with codecs.open(phfn, 'w', 'utf8') as phf:
for phone in phoneset:
phf.write(u'%s\n' % phone)
phf.write(u'SIL\n')
if ENABLE_NOISE_FILLER:
phf.write(u'NSPC\n')
logging.info("%s written." % phfn)
misc.render_template('data/src/speech/sphinx-run.sh.template',
'%s/sphinx-run.sh' % work_dir,
lm_name=lm_name)
misc.mkdirs('%s/local/dict.adapt' % data_dir)
dictfn2 = '%s/local/dict.adapt/lexicon.txt' % data_dir
logging.info ( "Exporting dictionary..." )
ps = {}
with open (dictfn2, 'w') as dictf:
dictf.write('!SIL SIL\n')
for token in sorted(lex):
ipa = lex[token]['ipa']
xsr = ipa2xsampa (token, ipa, spaces=True)
xs = xsr.replace('-','').replace('\' ', '\'').replace(' ', ' ').replace('#', 'nC')
dictf.write((u'%s %s\n' % (token, xs)).encode('utf8'))
for p in xs.split(' '):
if len(p)<1:
logging.error ( u"****ERROR: empty phoneme in : '%s' ('%s', ipa: '%s', token: '%s')" % (xs, xsr, ipa, token) )
pws = p[1:] if p[0] == '\'' else p
if not pws in ps:
ps[pws] = set([p])
else:
def lex_edit(token):
global lex, lex_token, lex_entry, lex_base
lex_set_token (token)
while True:
try:
lex_paint_main()
c = stdscr.getch()
if c == ord('q'):
lex.save()
break
# generate de-mary
elif c == ord('g'):
tts.locale = 'de'
tts.engine = 'mary'
tts.voice = 'bits3'
ipas = tts.gen_ipa (lex_base)
tts.say_ipa(ipas)
lex_entry['ipa'] = ipas
# generate de-espeak
elif c == ord('h'):
tts.locale ='de'
tts.engine ='espeak'
tts.voice ='de'
ipas = tts.gen_ipa (lex_base)
lex_entry['ipa'] = ipas
tts.locale ='de'
tts.engine ='mary'
tts.voice ='bits3'
tts.say_ipa(ipas)
# generate en-mary
elif c == ord('l'):
tts.locale ='en-US'
tts.engine ='mary'
tts.voice ='cmu-rms-hsmm'
ipas = tts.gen_ipa (lex_base)
tts.say_ipa(ipas)
lex_entry['ipa'] = ipas
# generate fr-mary
elif c == ord('k'):
tts.locale ='fr'
tts.engine ='mary'
tts.voice ='upmc-pierre-hsmm'
ipas = tts.gen_ipa (lex_base)
tts.say_ipa(ipas)
lex_entry['ipa'] = ipas
# generate de-sequitur
elif c == ord('j'):
ipas = sequitur_gen_ipa (SEQUITUR_MODEL, lex_base)
tts.locale ='de'
tts.engine ='mary'
tts.voice ='bits3'
tts.say_ipa(ipas)
lex_entry['ipa'] = ipas
# speak de mary unitsel
elif c == ord('p'):
if len(lex_entry['ipa']) == 0:
continue
ipas = lex_entry['ipa']
tts.locale = 'de'
tts.engine = 'mary'
tts.voice = 'bits3'
tts.say_ipa(ipas)
# speak de mary hsmm
elif c == ord('o'):
if len(lex_entry['ipa']) == 0:
continue
ipas = lex_entry['ipa']
tts.locale = 'de'
tts.engine = 'mary'
tts.voice = 'dfki-pavoque-neutral-hsmm'
tts.say_ipa(ipas)
# speak fr mary hsmm
elif c == ord('i'):
if len(lex_entry['ipa']) == 0:
continue
ipas = lex_entry['ipa']
tts.locale = 'fr'
tts.engine = 'mary'
tts.voice = 'pierre-voice-hsmm'
tts.say_ipa(ipas)
# speak en mary hsmm
elif c == ord('u'):
ipas = lex_entry['ipa']
tts.locale = 'en-US'
tts.engine = 'mary'
tts.voice = 'cmu-rms-hsmm'
tts.say_ipa(ipas)
# edit token
elif c == ord('t'):
token = misc.edit_popup(stdscr, ' Token ', '')
lex_set_token (token)
# edit XS
elif c == ord('e'):
ipas = lex_entry['ipa']
xs = ipa2xsampa (lex_token, ipas, stress_to_vowels=False)
xs = misc.edit_popup(stdscr, ' X-SAMPA ', xs)
ipas = xsampa2ipa (lex_token, xs)
lex_entry['ipa'] = ipas
except:
logging.error('EXCEPTION CAUGHT %s' % traceback.format_exc())
def lex_edit(token):
global lex, lex_token, lex_entry, lex_base
lex_set_token(token)
while True:
try:
lex_paint_main()
c = raw_input('lex > ').lower()
if c == 'q':
lex.save()
break
# generate de-mary
elif c == 'g':
tts.locale = 'de'
tts.engine = 'mary'
tts.voice = 'bits3'
ipas = tts.gen_ipa(lex_base)
tts.say_ipa(ipas)
lex_entry['ipa'] = ipas
# generate de-espeak
elif c == 'h':
tts.locale = 'de'
tts.engine = 'espeak'
tts.voice = 'de'
ipas = tts.gen_ipa(lex_base)
lex_entry['ipa'] = ipas
tts.locale = 'de'
tts.engine = 'mary'
tts.voice = 'bits3'
tts.say_ipa(ipas)
# generate en-mary
elif c == 'l':
tts.locale = 'en-US'
tts.engine = 'mary'
tts.voice = 'cmu-rms-hsmm'
ipas = tts.gen_ipa(lex_base)
tts.say_ipa(ipas)
lex_entry['ipa'] = ipas
# generate fr-mary
elif c == 'k':
tts.locale = 'fr'
tts.engine = 'mary'
tts.voice = 'upmc-pierre-hsmm'
ipas = tts.gen_ipa(lex_base)
tts.say_ipa(ipas)
lex_entry['ipa'] = ipas
# generate de-sequitur
elif c == 'j':
ipas = sequitur_gen_ipa(SEQUITUR_MODEL, lex_base)
tts.locale = 'de'
tts.engine = 'mary'
tts.voice = 'bits3'
tts.say_ipa(ipas)
lex_entry['ipa'] = ipas
# speak de mary unitsel
elif c == 'p':
if len(lex_entry['ipa']) == 0:
continue
ipas = lex_entry['ipa']
tts.locale = 'de'
tts.engine = 'mary'
tts.voice = 'bits3'
tts.say_ipa(ipas)
# speak de mary hsmm
elif c == 'o':
if len(lex_entry['ipa']) == 0:
continue
ipas = lex_entry['ipa']
tts.locale = 'de'
tts.engine = 'mary'
tts.voice = 'dfki-pavoque-neutral-hsmm'
tts.say_ipa(ipas)
# speak fr mary hsmm
elif c == 'i':
if len(lex_entry['ipa']) == 0:
continue
ipas = lex_entry['ipa']
tts.locale = 'fr'
tts.engine = 'mary'
tts.voice = 'pierre-voice-hsmm'
tts.say_ipa(ipas)
# speak en mary hsmm
elif c == 'u':
ipas = lex_entry['ipa']
tts.locale = 'en-US'
tts.engine = 'mary'
tts.voice = 'cmu-rms-hsmm'
tts.say_ipa(ipas)
# edit token
elif c == 't':
readline.add_history(lex_token.encode('utf8'))
token = raw_input('token: ').decode('utf8')
lex_set_token(token)
# edit XS
elif c == 'e':
ipas = lex_entry['ipa']
xs = ipa2xsampa(lex_token, ipas, stress_to_vowels=False)
readline.add_history(xs.encode('utf8'))
xs = raw_input('X-SAMPA: ').decode('utf8')
ipas = xsampa2ipa(lex_token, xs)
lex_entry['ipa'] = ipas
except:
logging.error('EXCEPTION CAUGHT %s' % traceback.format_exc())
def export_audio(train_val, tsdict):
global data_dir, utt_num, options, cnt
destdirfn = data_dir
lcnt = 0
for utt_id in tsdict:
ts = tsdict[utt_id]
tokens = tokenize(ts['ts'], lang=options.lang)
covered_by_lex = True
for token in tokens:
if not (token in lex):
logging.error(u'token %s missing from dict!' % token)
logging.error(u'utt_id: %s' % utt_id)
logging.error(u'ts: %s' % ts['ts'])
covered_by_lex = False
break
if not covered_by_lex:
continue
print("about to export")
with codecs.open('%s/%09d.id' % (destdirfn, utt_num[train_val]), 'w', 'utf8') as idf, \
codecs.open('%s/%09d.tkn' % (destdirfn, utt_num[train_val]), 'w', 'utf8') as tknf, \
codecs.open('%s/%09d.wrd' % (destdirfn, utt_num[train_val]), 'w', 'utf8') as wrdf :
print("utterance read. " + str(wrdf) + "\n")
tkn = u''
wrd = u''
for token in tokens:
ipas = lex[token]['ipa']
xsr = ipa2xsampa(token, ipas, spaces=True)
xs = (xsr.replace('-', '').replace('\' ', '\'').replace(
' ', ' ').replace('#', 'nC'))
if tkn:
tkn += u' | '
wrd += u' '
tkn += xs
wrd += token
tknf.write('%s\n' % tkn)
wrdf.write('%s\n' % wrd)
idf.write('utt_id\t%s\ncorpus\t%s\nlang\t%s\n' %
(utt_id, ts['corpus_name'], options.lang))
os.symlink('%s/%s/%s.wav' % (work_dir, ts['corpus_name'], utt_id),
'%s/%09d.wav' % (destdirfn, utt_num[train_val]))
# cmd = 'ln -s %s/%s/%s.wav %s/%09d.wav' % (wav16_dir, ts['corpus_name'], utt_id, destdirfn, utt_num[train_val])
# logging.debug(cmd)
# os.system(cmd)
# utt2spkf.write('%s %s\n' % (utt_id, ts['spk']))
utt_num[train_val] = utt_num[train_val] + 1
cnt += 1
lcnt += 1
if cnt % 1000 == 0:
logging.info(
'%6d audio files linked from %s [%s] (%6d/%6d)...' %
(cnt, ts['corpus_name'], train_val, lcnt, len(tsdict)))
tts.say_ipa(ipas, async=True)
# edit token
elif c == ord('t'):
token = misc.edit_popup(stdscr, ' Token ', '')
lex_set_token(token)
# edit XS
elif c == ord('e'):
ipas = lex_entry['ipa']
xs = ipa2xsampa(lex_token, ipas, stress_to_vowels=False)
xs = misc.edit_popup(stdscr, ' X-SAMPA ', xs)
try:
ipas = xsampa2ipa(lex_token, xs)
lex_entry['ipa'] = ipas
except:
pass
#
# fini
#
def synthesize(self, txt, mode='txt'):
if self._host_tts == 'local':
# import pdb; pdb.set_trace()
wav = None
if self.engine == 'mary':
self.marytts.voice = self._voice
self.marytts.locale = self._locale
if mode == 'txt':
wav = self.marytts.synth_wav(txt)
elif mode == 'ipa':
xs = ipa2mary('ipa', txt)
wav = self.marytts.synth_wav(xs, fmt='xs')
else:
raise Exception("unknown mary mode '%s'" % mode)
elif self.engine == 'espeak':
if mode == 'txt':
self.espeak.voice = self._voice
self.espeak.speed = self._speed
self.espeak.pitch = self._pitch
wav = self.espeak.synth_wav(txt)
# logging.debug ('synthesize: %s %s -> %s' % (txt, mode, repr(wav)))
elif mode == 'ipa':
xs = ipa2xsampa('ipa', txt)
logging.debug('synthesize: %s %s -> %s' %
(txt, mode, repr(xs)))
wav = self.espeak.synth_wav(xs, fmt='xs')
elif self.engine == 'pico':
if mode == 'txt':
self.picotts.voice = self._voice
wav = self.picotts.synth_wav(txt)
# logging.debug ('synthesize: %s %s -> %s' % (txt, mode, repr(wav)))
else:
raise Exception("unknown espeak mode '%s'" % mode)
else:
raise Exception("unknown engine '%s'" % self.engine)
else:
args = {
'l': self._locale,
'v': self._voice,
'e': self._engine,
'm': mode,
't': txt.encode('utf8')
}
url = 'http://%s:%s/tts/synth?%s' % (
self._host_tts, self._port_tts, urllib.urlencode(args))
response = requests.get(url)
if response.status_code != 200:
return None
wav = response.content
if wav:
logging.debug('synthesize: %s %s -> WAV' % (txt, mode))
else:
logging.error('synthesize: %s %s -> NO WAV' % (txt, mode))
return wav
