def config_stt(cache_dir, keywords, kws_last_modification_time_in_sec = None):
_ = Core()
cache_path = lambda x: os.path.join(cache_dir, x)
_.lang_file = cache_path('lm')
_.fsg_file = None
_.dic_file = cache_path('dic')
_.hmm_dir = cache_path('hmm')
# Save keywords for further pattern matching. Call `config_stt()` when switching minds to check for command file changes.
_.kwords = {}
_.max_w_cnt = 3
for phrase in keywords:
spl_ph = phrase.strip().split(' ')
w_cnt = len(spl_ph)
if w_cnt > _.max_w_cnt:
_.max_w_cnt = w_cnt
for kword in spl_ph:
# Combine all phrases which use keywords.
r_phrases = _.kwords.setdefault(kword,{})
r_phrases[phrase] = w_cnt
# Save phrases.
_.phrases = [x.strip().replace('%d', '').upper() for x in sorted(keywords)]
# Check if commands file was modified.
if kws_last_modification_time_in_sec:
if os.path.exists(_.dic_file) and (kws_last_modification_time_in_sec < stat_mtime(_.dic_file)):
return _
_.strings_file = cache_path("sentences.corpus")
data = '\n'.join(_.phrases)
write_file(_.strings_file, data)
# Download language model data from `speech.cs.cmu.edu`.
update_language(_)
return _
def get_decoder():
# XXX: race condition when mind isn't set yet
mind = oa.core.mind
if not hasattr(_decoders, mind.name):
# Configure Speech to text dictionaries.
ret = config_stt(mind.cache_dir, mind.kws.keys(), stat_mtime(mind.module))
# Process audio chunk by chunk. On a keyphrase detected perform the action and restart search.
config = Decoder.default_config()
# Set paths for the language model files.
config.set_string('-hmm', ret.hmm_dir)
config.set_string("-lm", ret.lang_file)
config.set_string("-dict", ret.dic_file)
config.set_string("-logfn", os.devnull) # Disable logging.
ret.decoder = Decoder(config)
_decoders[mind.name] = ret
else:
return _decoders[mind.name]
return ret