def google_STT(self,audio):
client = speech_v1.SpeechClient.from_service_account_json(
'/data/second-conquest-293723-05738e995f8f.json')
# Loads the audio into memory
with io.open(audio, "rb") as audio_file:
content = audio_file.read()
audio = speech_v1.RecognitionAudio(content=content)
encoding = speech_v1.RecognitionConfig.AudioEncoding.ENCODING_UNSPECIFIED
config = speech_v1.RecognitionConfig(
encoding=encoding,
sample_rate_hertz=22050,
language_code="en-US",
enable_automatic_punctuation=True,
)
# Detects speech in the audio file
start = time.time()
response = client.recognize(request={"config": config, "audio": audio})
text = ''
for result in response.results:
text = text + result.alternatives[0].transcript
text = english_cleaners(text)
return text
def make_arpabet(text):
# g2p functions
g2p = G2p()
# Define punctuation, prevent punctuation curlies, and make replacement dictionary to fix spacing
punc = "!?,.;:#-_'\"()[]\n,."
punc = list(punc)
punc_key = list(punc)
punc_alt = [" " + item for item in punc]
punc_dict = {}
for key in punc_alt:
for value in punc_key:
punc_dict[key] = value
punc_key.remove(value)
break
# Text processing
text = " ".join(g2p(english_cleaners(text))).split(" ")
outlist = []
for item in text:
item = item.strip()
if item not in punc:
item = "{" + item + "}"
outlist.append(item)
text = " ".join(outlist)
for key, replacement in punc_dict.items():
text = text.replace(key, replacement)
return text
def convert_to_ipa(texts):
print("Converting training files to IPA notation...")
epi = epitran.Epitran('eng-Latn', ligatures=True)
for text_mel_pair in texts:
text_mel_pair[1] = ipa.convert(english_cleaners(text_mel_pair[1]))
foreign_words = re.findall(r"[^ ]{0,}\*", text_mel_pair[1])
for word in foreign_words:
text_mel_pair[1] = text_mel_pair[1].replace(
word, epi.transliterate(word[0:len(word) - 1]))
def generate_from_file(tacotron2_path, waveglow_path, text_file, output_directory):
# Make synthesis paths
if not os.path.exists(output_directory):
os.makedirs(output_directory)
print("Creating directory " + output_directory + "...")
hparams = create_hparams()
hparams.sampling_rate = 22050
print("Loading models...")
model = load_model(hparams)
model.load_state_dict(torch.load(tacotron2_path)['state_dict'])
_ = model.cuda().eval().half()
waveglow = torch.load(waveglow_path)['model']
waveglow.cuda().eval().half()
for k in waveglow.convinv:
k.float()
denoiser = Denoiser(waveglow)
genlist = []
with open(text_file) as file:
for line in file:
genlist.append(line.strip())
for entry in genlist:
wav_name = "_".join(entry.split(" ")[:4]).lower() + ".wav"
epi = epitran.Epitran('eng-Latn', ligatures = True)
if hparams.preprocessing == "ipa":
entry = ipa.convert(english_cleaners(entry))
foreign_words = re.findall(r"[^ ]{0,}\*", entry)
for word in foreign_words:
entry = entry.replace(word, epi.transliterate(word[0:len(word)-1]))
if hparams.preprocessing == "arpabet":
entry = make_arpabet(entry)
# Text sequencer
if hparams.preprocessing is not None:
sequence = np.array(text_to_sequence(entry, None))[None, :]
else:
sequence = np.array(text_to_sequence(entry, ['english_cleaners']))[None, :]
sequence = torch.autograd.Variable(
torch.from_numpy(sequence)).cuda().long()
# Synthesis
mel_outputs, mel_outputs_postnet, _, alignments = model.inference(sequence)
with torch.no_grad():
audio = waveglow.infer(mel_outputs_postnet, sigma=0.666)
audio_denoised = denoiser(audio, strength=0.01)[:, 0]
# Save audio
print ("Saving " + wav_name)
write(os.path.join(output_directory, wav_name), hparams.sampling_rate, audio_denoised[0].data.cpu().numpy())
def get_first_words_idx(script_path):
with open(script_path) as f:
text = f.read()
sents = sent_tokenize(text)
print('sentences number: ', len(sents))
text = english_cleaners(text)
words = text.split(' ')
first_words_idx = [0]
# 문장 마지막 단어 인덱스 구하기
for idx, word in enumerate(words):
if word.endswith('.') or word.endswith('?') or word.endswith('!'):
first_words_idx.append(idx + 1)
return first_words_idx
def read_csv(path, fn_encoding='UTF8'):
# reads csv file into audio snippet name and its transcript
with open(path, encoding=fn_encoding) as f:
data = []
temp = '' # for storing non-normalized
for line in f:
audio_name, audio_transcript, deprecated_1 = line.split('\t')
replace_audio_name = audio_name.replace("/", "\\")
audio_transcript = audio_transcript.strip()
audio_transcript = audio_transcript.replace("\"", "").replace("(", "").replace(")", "")\
.replace("[", "").replace("]", "")
audio_normalized_transcript = english_cleaners(audio_transcript)
data.append(
Data(replace_audio_name, audio_transcript,
audio_normalized_transcript))
return data
def MFA_file_prep(self, hparams):
''' Will Create .lab files for all existing wav files in the hparams.wav_path folder.
The .lab files will contain the cleaned or normalized text (expanded abbreviations, remove punctuations, expand number, to upper case) from csv_file
The function will also check if the words in content are there in dict or not
And will create a dictionary {alien_words: phoneme(alien_words)} using g2p_en
input - hparams
output - new_word_dictionary
'''
filenames = []
content = []
update_words = {}
with open(hparams.csv_path, encoding='utf-8') as f:
for lines in f:
filenames.append(lines.split("|")[0])
content.append(lines.split('|')[1])
words = self.load_words_from_dict(hparams)
for i in range(0, len(filenames)):
if os.path.exists(f'{hparams.lab_path}/{filenames[i]}.wav'):
path = os.path.join(hparams.lab_path, filenames[i] + ".lab")
clean_content = english_cleaners(content[i])
clean_content = punctuation_removers(
clean_content) # add remove punctuations
f = open(path, 'w+')
f.write(clean_content.upper())
f.close()
alien = set(clean_content.upper().split()) - set(words)
alien_update = {i: (g2p(i)) for i in list(alien)}
update_words = {**update_words, **alien_update}
if update_words:
print("update your dictionary using update_dict() of this class")
else:
print("No dictionary update required")
return update_words #words to update
def test_cleaner_pipelines():
text = 'Mr. Müller ate 2 Apples'
assert cleaners.english_cleaners(text) == 'mister muller ate two apples'
assert cleaners.transliteration_cleaners(text) == 'mr. muller ate 2 apples'
assert cleaners.basic_cleaners(text) == 'mr. müller ate 2 apples'
def text_recognition(path, config):
root, ext = os.path.splitext(path)
txt_path = root + ".txt"
if os.path.exists(txt_path):
with open(txt_path) as f:
out = json.loads(open(txt_path).read())
return out
# if new api account is used, do resetting env file for google credential
from google.cloud import speech
from google.cloud.speech import enums
from google.cloud.speech import types
out = {}
error_count = 0
tmp_path = os.path.splitext(path)[0] + ".wav"
client = speech.SpeechClient() # Fixed
while True:
try:
# client= speech.SpeechClient() # Causes 10060 max retries exceeded -to OAuth -HK
content = path[0]
with io.open(tmp_path, 'rb') as f:
audio = types.RecognitionAudio(content=f.read())
config = types.RecognitionConfig(
encoding=enums.RecognitionConfig.AudioEncoding.LINEAR16,
sample_rate_hertz=16000,
language_code='en-GB')
response = client.recognize(config, audio)
if len(response.results) > 0:
alternatives = response.results[0].alternatives
#results = 실제 음성인식된 text
results = [
alternative.transcript for alternative in alternatives
]
assert len(results) == 1, "More than 1 results: {}".format(
results)
#실질적으로 txt
out = {
os.path.basename(path):
"" if len(results) == 0 else results[0],
"normalized_text": english_cleaners(results[0])
}
print(path, results[0], english_cleaners(results[0]))
break
break
except Exception as err:
raise Exception("OS error: {0}".format(err))
error_count += 1
print("Skip warning for {} for {} times".format(path, error_count))
if error_count > 5:
break
else:
continue
global removed
if len(out) is 0:
# remove file that only has instrument sound.
os.remove(root + '.wav')
print(root, '.wav file is removed!')
else:
with open(txt_path, 'w') as f:
json.dump(out, f, indent=2, ensure_ascii=False)
return out
def english_cleaner():
actual = cleaners.english_cleaners(
'I want to be there early on the day. Please organize')
expected = 'I want to be there early on the day. Please organize'
assert actual == expected
