def __init__(self, hmm='data/spanish/CIEMPIESS_Spanish_Models_581h/Models/modelo',
lm='data/spanish/CIEMPIESS_Spanish_Models_581h/Models/leng.lm.bin',
dict='data/spanish/CIEMPIESS_Spanish_Models_581h/Models/dicc.dic',
grammar='data/gramatica-tp2.gram', dataPath='tmp/'):
self.data_path = dataPath
config = {
'hmm': hmm,
'lm': lm,
'dict': dict
}
#model_path = get_model_path()
self.ps = Pocketsphinx(**config)
# Switch to JSGF grammar
jsgf = Jsgf(grammar)
rule = jsgf.get_rule('tp2.grammar')
fsg = jsgf.build_fsg(rule, self.ps.get_logmath(), 7.5)
self.ps.set_fsg('tp2', fsg)
self.ps.set_search('tp2')
# Síntesis
self.tts_authenticator = IAMAuthenticator('cq9_4YcCXxClw2AfgUhbokFktZ-xSRT4kcHS2akcZ05J')
self.tts = TextToSpeechV1(authenticator=self.tts_authenticator)
self.tts.set_service_url('https://stream.watsonplatform.net/text-to-speech/api')
def recognize(self, audio_data, keyword_entries=None, grammar=None):
language = self.lang
assert isinstance(audio_data,
AudioData), "``audio_data`` must be audio data"
assert isinstance(language, str), "``language`` must be a string"
assert keyword_entries is None or all(
isinstance(keyword,
(type(""), type(u""))) and 0 <= sensitivity <= 1
for keyword, sensitivity in
keyword_entries), "``keyword_entries`` must be ``None`` or" \
" a list of pairs of strings and " \
"numbers between 0 and 1"
# obtain audio data
raw_data = audio_data.get_raw_data(convert_rate=16000, convert_width=2)
# obtain recognition results
if keyword_entries is not None: # explicitly specified set of keywords
with PortableNamedTemporaryFile("w") as f:
# generate a keywords file
f.writelines("{} /1e{}/\n".format(keyword, 100 * sensitivity -
110)
for keyword, sensitivity in keyword_entries)
f.flush()
# perform the speech recognition with the keywords file
self.decoder.set_kws("keywords", f.name)
self.decoder.set_search("keywords")
self.decoder.start_utt() # begin utterance processing
self.decoder.process_raw(raw_data, False, True)
self.decoder.end_utt() # stop utterance processing
elif grammar is not None: # a path to a FSG or JSGF grammar
if not os.path.exists(grammar):
raise ValueError(
"Grammar '{0}' does not exist.".format(grammar))
grammar_path = os.path.abspath(os.path.dirname(grammar))
grammar_name = os.path.splitext(os.path.basename(grammar))[0]
fsg_path = "{0}/{1}.fsg".format(grammar_path, grammar_name)
if not os.path.exists(
fsg_path): # create FSG grammar if not available
jsgf = Jsgf(grammar)
rule = jsgf.get_rule("{0}.{0}".format(grammar_name))
fsg = jsgf.build_fsg(rule, self.decoder.get_logmath(), 7.5)
fsg.writefile(fsg_path)
else:
fsg = FsgModel(fsg_path, self.decoder.get_logmath(), 7.5)
self.decoder.set_fsg(grammar_name, fsg)
self.decoder.set_search(grammar_name)
self.decoder.start_utt()
self.decoder.process_raw(raw_data, False, True)
self.decoder.end_utt() # stop utterance processing
else: # no keywords, perform freeform recognition
self.decoder.start_utt() # begin utterance processing
self.decoder.process_raw(raw_data, False, True)
self.decoder.end_utt() # stop utterance processing
# return results
hypothesis = self.decoder.hyp()
if hypothesis is not None:
return hypothesis.hypstr
raise UnknownValueError() # no transcriptions available
def retrieve_scores(word):
filename = word + '.wav'
grammarname = word + '-align.jsgf'
model_path = get_model_path()
# Initialize the config values
config = DefaultConfig()
config.set_boolean('-verbose', False)
config.set_string('-hmm', os.path.join(model_path, 'en-us'))
config.set_boolean('-lm', False)
config.set_string('-dict', 'phonemes.dict.txt')
config.set_boolean('-backtrace', True)
config.set_boolean('-bestpath', False)
config.set_boolean('-fsgusefiller', False)
decoder = Decoder(config)
# Set the search to JSGF Grammar
jsgf = Jsgf(grammarname)
rule = jsgf.get_rule('forcing.' + word)
decoder.set_jsgf_file('grammar', grammarname)
decoder.set_search('grammar')
stream = open(filename, 'rb')
utt_started = False
scores = []
decoder.start_utt()
while True:
buf = stream.read(1024)
if buf:
decoder.process_raw(buf, False, False)
in_speech = decoder.get_in_speech()
if (in_speech and not utt_started):
utt_started = True
if (not in_speech and utt_started):
decoder.end_utt()
hyp = decoder.hyp()
if hyp is not None:
print('hyp: %s' % (hyp.best_score))
print_segments(decoder)
scores = retrieve_segments(decoder)
decoder.start_utt()
utt_started = False
else:
break
decoder.end_utt()
print('scores:', scores)
return scores
def test_jsgf(self):
ps = Pocketsphinx(lm='deps/pocketsphinx/test/data/turtle.lm.bin',
dic='deps/pocketsphinx/test/data/turtle.dic')
# Decoding with 'turtle' language model
ps.decode()
self.assertEqual(ps.hypothesis(), 'go forward ten meters')
# Switch to JSGF grammar
jsgf = Jsgf('deps/pocketsphinx/test/data/goforward.gram')
rule = jsgf.get_rule('goforward.move2')
fsg = jsgf.build_fsg(rule, ps.get_logmath(), 7.5)
ps.set_fsg('goforward', fsg)
ps.set_search('goforward')
# Decoding with 'goforward' grammar
ps.decode()
self.assertEqual(ps.hypothesis(), 'go forward ten meters')
def test_jsgf(self):
ps = Pocketsphinx(
lm='deps/pocketsphinx/test/data/turtle.lm.bin',
dic='deps/pocketsphinx/test/data/turtle.dic'
)
# Decoding with 'turtle' language model
ps.decode()
self.assertEqual(ps.hypothesis(), 'go forward ten meters')
# Switch to JSGF grammar
jsgf = Jsgf('deps/pocketsphinx/test/data/goforward.gram')
rule = jsgf.get_rule('goforward.move2')
fsg = jsgf.build_fsg(rule, ps.get_logmath(), 7.5)
ps.set_fsg('goforward', fsg)
ps.set_search('goforward')
# Decoding with 'goforward' grammar
ps.decode()
self.assertEqual(ps.hypothesis(), 'go forward ten meters')
def recognize_sphinx(self, audio_data, language="en-US", keyword_entries=None, grammar=None, show_all=False):
"""
Performs speech recognition on ``audio_data`` (an ``AudioData`` instance), using CMU Sphinx.
The recognition language is determined by ``language``, an RFC5646 language tag like ``"en-US"`` or ``"en-GB"``, defaulting to US English. Out of the box, only ``en-US`` is supported. See `Notes on using `PocketSphinx <https://github.com/Uberi/speech_recognition/blob/master/reference/pocketsphinx.rst>`__ for information about installing other languages. This document is also included under ``reference/pocketsphinx.rst``. The ``language`` parameter can also be a tuple of filesystem paths, of the form ``(acoustic_parameters_directory, language_model_file, phoneme_dictionary_file)`` - this allows you to load arbitrary Sphinx models.
If specified, the keywords to search for are determined by ``keyword_entries``, an iterable of tuples of the form ``(keyword, sensitivity)``, where ``keyword`` is a phrase, and ``sensitivity`` is how sensitive to this phrase the recognizer should be, on a scale of 0 (very insensitive, more false negatives) to 1 (very sensitive, more false positives) inclusive. If not specified or ``None``, no keywords are used and Sphinx will simply transcribe whatever words it recognizes. Specifying ``keyword_entries`` is more accurate than just looking for those same keywords in non-keyword-based transcriptions, because Sphinx knows specifically what sounds to look for.
Sphinx can also handle FSG or JSGF grammars. The parameter ``grammar`` expects a path to the grammar file. Note that if a JSGF grammar is passed, an FSG grammar will be created at the same location to speed up execution in the next run. If ``keyword_entries`` are passed, content of ``grammar`` will be ignored.
Returns the most likely transcription if ``show_all`` is false (the default). Otherwise, returns the Sphinx ``pocketsphinx.pocketsphinx.Decoder`` object resulting from the recognition.
Raises a ``speech_recognition.UnknownValueError`` exception if the speech is unintelligible. Raises a ``speech_recognition.RequestError`` exception if there are any issues with the Sphinx installation.
"""
assert isinstance(audio_data, AudioData), "``audio_data`` must be audio data"
assert isinstance(language, str) or (isinstance(language, tuple) and len(language) == 3), "``language`` must be a string or 3-tuple of Sphinx data file paths of the form ``(acoustic_parameters, language_model, phoneme_dictionary)``"
assert keyword_entries is None or all(isinstance(keyword, (type(""), type(u""))) and 0 <= sensitivity <= 1 for keyword, sensitivity in keyword_entries), "``keyword_entries`` must be ``None`` or a list of pairs of strings and numbers between 0 and 1"
# import the PocketSphinx speech recognition module
try:
from pocketsphinx import pocketsphinx, Jsgf, FsgModel
except ImportError:
raise RequestError("missing PocketSphinx module: ensure that PocketSphinx is set up correctly.")
except ValueError:
raise RequestError("bad PocketSphinx installation; try reinstalling PocketSphinx version 0.0.9 or better.")
if not hasattr(pocketsphinx, "Decoder") or not hasattr(pocketsphinx.Decoder, "default_config"):
raise RequestError("outdated PocketSphinx installation; ensure you have PocketSphinx version 0.0.9 or better.")
if isinstance(language, str): # directory containing language data
language_directory = os.path.join(os.path.dirname(os.path.realpath(__file__)), "pocketsphinx-data", language)
if not os.path.isdir(language_directory):
raise RequestError("missing PocketSphinx language data directory: \"{}\"".format(language_directory))
acoustic_parameters_directory = os.path.join(language_directory, "acoustic-model")
language_model_file = os.path.join(language_directory, "language-model.lm.bin")
phoneme_dictionary_file = os.path.join(language_directory, "pronounciation-dictionary.dict")
else: # 3-tuple of Sphinx data file paths
acoustic_parameters_directory, language_model_file, phoneme_dictionary_file = language
if not os.path.isdir(acoustic_parameters_directory):
raise RequestError("missing PocketSphinx language model parameters directory: \"{}\"".format(acoustic_parameters_directory))
if not os.path.isfile(language_model_file):
raise RequestError("missing PocketSphinx language model file: \"{}\"".format(language_model_file))
if not os.path.isfile(phoneme_dictionary_file):
raise RequestError("missing PocketSphinx phoneme dictionary file: \"{}\"".format(phoneme_dictionary_file))
# create decoder object
config = pocketsphinx.Decoder.default_config()
config.set_string("-hmm", acoustic_parameters_directory) # set the path of the hidden Markov model (HMM) parameter files
config.set_string("-lm", language_model_file)
config.set_string("-dict", phoneme_dictionary_file)
config.set_string("-logfn", os.devnull) # disable logging (logging causes unwanted output in terminal)
decoder = pocketsphinx.Decoder(config)
# obtain audio data
raw_data = audio_data.get_raw_data(convert_rate=16000, convert_width=2) # the included language models require audio to be 16-bit mono 16 kHz in little-endian format
# obtain recognition results
if keyword_entries is not None: # explicitly specified set of keywords
with PortableNamedTemporaryFile("w") as f:
# generate a keywords file - Sphinx documentation recommendeds sensitivities between 1e-50 and 1e-5
f.writelines("{} /1e{}/\n".format(keyword, 100 * sensitivity - 110) for keyword, sensitivity in keyword_entries)
f.flush()
# perform the speech recognition with the keywords file (this is inside the context manager so the file isn;t deleted until we're done)
decoder.set_kws("keywords", f.name)
decoder.set_search("keywords")
decoder.start_utt() # begin utterance processing
decoder.process_raw(raw_data, False, True) # process audio data with recognition enabled (no_search = False), as a full utterance (full_utt = True)
decoder.end_utt() # stop utterance processing
elif grammar is not None: # a path to a FSG or JSGF grammar
if not os.path.exists(grammar):
raise ValueError("Grammar '{0}' does not exist.".format(grammar))
grammar_path = os.path.abspath(os.path.dirname(grammar))
grammar_name = os.path.splitext(os.path.basename(grammar))[0]
fsg_path = "{0}/{1}.fsg".format(grammar_path, grammar_name)
if not os.path.exists(fsg_path): # create FSG grammar if not available
jsgf = Jsgf(grammar)
rule = jsgf.get_rule("{0}.{0}".format(grammar_name))
fsg = jsgf.build_fsg(rule, decoder.get_logmath(), 7.5)
fsg.writefile(fsg_path)
else:
fsg = FsgModel(fsg_path, decoder.get_logmath(), 7.5)
decoder.set_fsg(grammar_name, fsg)
decoder.set_search(grammar_name)
decoder.start_utt()
decoder.process_raw(raw_data, False, True) # process audio data with recognition enabled (no_search = False), as a full utterance (full_utt = True)
decoder.end_utt() # stop utterance processing
else: # no keywords, perform freeform recognition
decoder.start_utt() # begin utterance processing
decoder.process_raw(raw_data, False, True) # process audio data with recognition enabled (no_search = False), as a full utterance (full_utt = True)
decoder.end_utt() # stop utterance processing
if show_all: return decoder
# return results
hypothesis = decoder.hyp()
if hypothesis is not None: return hypothesis.hypstr
raise UnknownValueError() # no transcriptions available
config = nu_config()
r = sr.Recognizer()
language = config.get('language', 'default')
model_dir = os.getcwd(
) + '/nu/modules/brain/senses/text2speech/languages/' + language
grammar_path = model_dir + '/grammars'
config = pocketsphinx.Decoder.default_config()
config.set_string('-hmm', model_dir + '/accoustic-model')
config.set_string('-lm', model_dir + '/language-model.bin')
config.set_string('-dict', model_dir + '/pronounciation-dictionary.dict')
config.set_string("-logfn", os.devnull)
jsgf = Jsgf(grammar_path)
grammar_decoders = []
pattern = re.compile('public <(.*?)> =')
with open(grammar_path, 'rt') as in_file:
for linenum, line in enumerate(in_file):
grammar_key = pattern.findall(line)
if grammar_key != []:
decoder = pocketsphinx.Decoder(config)
ruleGrammar = jsgf.get_rule(
('structure.' + grammar_key[0]).format(grammar_path))
fsgNext = jsgf.build_fsg(ruleGrammar, decoder.get_logmath(), 7.5)
decoder.set_fsg(grammar_key[0], fsgNext)
decoder.set_search(grammar_key[0])
grammar_decoders.append(decoder)
def __init__(self, **kwargs):
signal.signal(signal.SIGINT, self.stop)
model_path = get_model_path()
kwargs = {
x: os.path.expandvars(kwargs[x])
if type(kwargs[x]) is str else kwargs[x]
for x in kwargs
}
nodename = kwargs.pop('nodename')
grammar_file = kwargs.pop('grammar_file', None)
grammar_rule = kwargs.pop('grammar_rule', None)
grammar_name = kwargs.pop('grammar_name', None)
kwargs.pop('esiaf_input_topic')
if kwargs.get('dic') is not None and kwargs.get('dict') is None:
kwargs['dict'] = kwargs.pop('dic')
if kwargs.get('hmm') is None:
kwargs['hmm'] = os.path.join(model_path, 'en-us')
if kwargs.get('lm') is None:
kwargs['lm'] = os.path.join(model_path, 'en-us.lm.bin')
if kwargs.get('dict') is None and kwargs.get('dic') is None:
kwargs['dict'] = os.path.join(model_path, 'cmudict-en-us.dict')
if kwargs.pop('verbose', False) is False:
if sys.platform.startswith('win'):
kwargs['logfn'] = 'nul'
else:
kwargs['logfn'] = '/dev/null'
config = Decoder.default_config()
print(kwargs)
for key, value in kwargs.items():
if isinstance(value, bool):
config.set_boolean('-{}'.format(key), value)
elif isinstance(value, int):
config.set_int('-{}'.format(key), value)
elif isinstance(value, float):
config.set_float('-{}'.format(key), value)
elif isinstance(value, str):
config.set_string('-{}'.format(key), value)
self.decoder = Decoder(config)
if grammar_file and grammar_rule and grammar_name:
jsgf = Jsgf(grammar_file)
rule = jsgf.get_rule(grammar_name + '.' + grammar_rule)
fsg = jsgf.build_fsg(rule, self.decoder.get_logmath(), 7.5)
self.decoder.set_fsg(grammar_name, fsg)
self.decoder.set_search(grammar_name)
self.start = None
self.finish = None
self.speech_publisher = rospy.Publisher(nodename + '/' + 'SpeechRec',
SpeechInfo,
queue_size=10)