def test_add_corpus(self):
corpus = TrainingCorpus()
nb = corpus.add_corpus( os.path.join(SAMPLES_PATH,"samples-eng") )
self.assertEqual(nb, 0)
nb = corpus.add_corpus( os.path.join(SAMPLES_PATH,"samples-ita") )
self.assertEqual(nb, 4)
corpus.add_corpus( SAMPLES_PATH )
self.assertGreater(len(corpus.transfiles), 10)
self.assertEqual(len(corpus.phonfiles), 0)
self.assertEqual(len(corpus.alignfiles), 0)
def test_trainer_with_data(self):
#setup_logging(1,None)
corpus = TrainingCorpus()
corpus.fix_resources(dictfile=os.path.join(RESOURCES_PATH, "dict", "fra.dict"), mappingfile=os.path.join(RESOURCES_PATH,"models","models-fra","monophones.repl" ))
corpus.lang = "fra"
corpus.datatrainer.protodir = os.path.join(HERE,"protos")
corpus.add_file( os.path.join(HERE,"F_F_B003-P8-palign.TextGrid"), os.path.join(HERE,"F_F_B003-P8.wav") )
corpus.add_file( os.path.join(HERE,"track_0001-phon.xra"), os.path.join(HERE,"track_0001.wav") )
corpus.add_corpus( os.path.join(SAMPLES_PATH,"samples-fra") )
trainer = HTKModelTrainer( corpus )
acmodel = trainer.training_recipe( delete=True )
# ---------------------------------
# 2. Create a Corpus Manager
# it manages the set of training data:
# - establishes the list of phonemes (from the dict);
# - converts the input annotated data into the HTK-specific data format;
# - codes the audio data.
corpus = TrainingCorpus( datatrainer, lang=args.l )
corpus.fix_resources( dictfile=args.r, mappingfile=args.m )
if args.i:
for entry in args.i:
if os.path.isdir( entry ):
corpus.add_corpus( entry )
else:
logging.info('[ WARNING ] Ignore the given entry: %s'%entry)
# ---------------------------------
# 3. Acoustic Model Training
trainer = HTKModelTrainer( corpus )
DELETE = False
if args.t is None:
DELETE = True
trainer.training_recipe( outdir=args.o, delete=DELETE )
# ---------------------------------------------------------------------------
