def g2p(self, word):
res = self.translator(word)
xs = u' '.join(res)
ipa = xsampa2ipa(word, xs)
return ipa
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 sequitur_gen_ipa_multi(modelfn, words):
ipa_map = {}
with tempfile.NamedTemporaryFile() as f:
for word in words:
f.write((u'%s\n' % word).encode('utf8'))
f.flush()
cmd = ['g2p.py', '--model', modelfn, '--apply', f.name]
res = misc.run_command(cmd, capture_stderr=False)
logging.debug('%s' % ' '.join(cmd))
for l in res:
line = l.strip()
logging.debug('%s' % line)
if 'stack usage:' in line:
continue
parts = line.decode('utf8', errors='ignore').split('\t')
if len(parts) < 2:
continue
try:
word = parts[0]
if word in words:
xs = parts[1]
# print 'XS', xs
ipa = xsampa2ipa(word, xs)
ipa_map[word] = ipa
except:
logging.error("Error processing line %s:" % line)
logging.error(traceback.format_exc())
return ipa_map
def sequitur_gen_ipa(modelfn, word):
ipa = u''
with tempfile.NamedTemporaryFile() as f:
f.write((u'%s\n' % word).encode('utf8'))
f.flush()
cmd = ['g2p.py', '--model', modelfn, '--apply', f.name]
res = misc.run_command(cmd)
logging.debug('%s' % ' '.join(cmd))
for l in res:
line = l.strip()
logging.debug('%s' % line)
if 'stack usage:' in line:
continue
if word in line.decode('utf8'):
parts = line.split('\t')
if len(parts) < 2:
continue
xs = parts[1]
# print 'XS', xs
ipa = xsampa2ipa(word, xs)
return ipa
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
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())
stress = False
if ph[l - 1] == u'0':
ph = ph[:l - 1]
elif ph[l - 1] == u'1':
ph = ph[:l - 1]
stress = True
elif ph[l - 1] == u'2':
ph = ph[:l - 1]
# stress = True
if stress:
xs += u"'"
xs += CMU2XS[ph] + u' '
ipa = xsampa2ipa(word, xs)
# logging.debug(u'%s %s %s' % (word, xs, ipa))
logging.debug(u'%s %s' % (word, ipa))
lex_new[word] = ipa
if DEBUG_LIMIT and len(lex_new) >= DEBUG_LIMIT:
logging.warn('DEBUG LIMIT REACHED!')
break
# #
# # diff against existing dict
# #
#
# print "loading lexicon..."
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
#
curses.nocbreak()
stdscr.keypad(0)
curses.echo()
curses.endwin()
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~")
res = xsampa2ipa(u"entrée A~ t R e", u"A~ t R e")
#print "res: %s" % res
self.assertEqual(res, u"ɑ̃tʁe")
xs = ''
for phone in parts[1:]:
ph = phone
if ph.endswith('0'):
# xs += "'"
ph = ph[0:len(ph) - 1]
if ph.endswith('1'):
xs += "'"
ph = ph[0:len(ph) - 1]
if ph.endswith('2'):
# xs += "'"
ph = ph[0:len(ph) - 1]
if ph in ab2xs:
xs += ab2xs[ph]
else:
logging.error('unknown phone: %s' % ph)
sys.exit(1)
ipa = phonetics.xsampa2ipa(word, xs)
# print line.strip(), ' -> ', word, xs
outf.write(u'%s;%s\n' % (word, ipa))
cnt += 1
logging.info('%d entries written to %s . skipped: %d' %
(cnt, OUTDICT, skipped))