MycroftAI
/
mimic2
mirror of https://github.com/MycroftAI/mimic2.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

abstracted synthesize_helper out

pull/15/head
Michael Nguyen 2018-07-16 15:23:10 -05:00
parent 3ea41b8820
commit 126ef34d44
4 changed files with 162 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

25
demo_server.py
View File

@ -3,13 +3,19 @@ from flask.views import MethodView
from hparams import hparams, hparams_debug_string
import argparse
import os
from util import audio
from synthesizer import Synthesizer
from flask_cors import CORS
import io
import numpy as np
import math
from synthesize_helper import synthesize_helper
app = Flask(__name__)
CORS(app)
use_synthesize_helper = False
html_body = '''<html><title>Demo</title>
<style>
body {padding: 16px; font-family: sans-serif; font-size: 14px; color: #444}
@ -63,14 +69,18 @@ function synthesize(text) {
synthesizer = Synthesizer()
class Mimic2(MethodView):
def get(self):
text = request.args.get('text')
if text:
wav, _ = synthesizer.synthesize(text)
audio = io.BytesIO(wav)
return send_file(audio, mimetype="audio/wav")
if use_synthesize_helper:
wav = synthesize_helper(text, synthesizer)
# wav, _ = synthesizer.synthesize(text)
audio = io.BytesIO(wav)
return send_file(audio, mimetype="audio/wav")
else:
wav, _ = synthesizer.synthesize(text)
return send_file(wav, mimetype="audio/wav")
class UI(MethodView):
@ -95,7 +105,14 @@ if __name__ == '__main__':
parser.add_argument(
'--gpu_assignment', default='0',
help='Set the gpu the model should run on')
parser.add_argument(
'--synthezier_helper', default=False, action="store_false",
help='uses the synthesize helper during sythesis'
)
args = parser.parse_args()
use_synthesize_helper = args.synthezier_helper
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu_assignment
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
hparams.parse(args.hparams)

6
export.py
View File

@ -27,7 +27,7 @@ if __name__ == "__main__":
)
wav_output = tf.saved_model.utils.build_tensor_info(synth.wav_output)
alignment = tf.saved_model.utils.build_tensor_info(synth.alignment)
# alignment = tf.saved_model.utils.build_tensor_info(synth.alignment)
prediction_signature = (
tf.saved_model.signature_def_utils.build_signature_def(
@ -36,8 +36,8 @@ if __name__ == "__main__":
"input_lengths": input_lengths
},
outputs={
'wav_output': wav_output,
'alignment': alignment
'wav_output': wav_output #,
# 'alignment': alignment
},
method_name=tf.saved_model.signature_constants.PREDICT_METHOD_NAME)
)

133
synthesize_helper.py Normal file
View File

@ -0,0 +1,133 @@
"""An optional module to post preprocess the text before syntehsizing
"""
import io
import math
from util import audio
import numpy as np
punctuations = ['.', '?', '!']
split_punctuations = [',', '.', '-', '?', '!', ':', ';']
letter_lookup = {
'A': 'ayy',
'B': 'bee',
'C': 'see',
'D': 'dee',
'E': 'eee',
'F': 'eff',
'G': 'jee',
'H': 'aitch',
'I': 'eye',
'J': 'jay',
'K': 'kay',
'L': 'el',
'M': 'em',
'N': 'en',
'O': 'oow',
'P': 'pee',
'Q': 'queue',
'R': 'are',
'S': 'es',
'T': 'tee',
'U': 'you',
'V': 'vee',
'W': 'double you',
'X': 'ex',
'Y': 'why',
'Z': 'zee'
}
def replace_acronym(text):
for idx, word in enumerate(text):
if len(word) == 1:
continue
if word.isupper():
sound = ""
for letter in word.strip():
if letter_lookup.get(letter):
sound += letter_lookup.get(letter) + " "
text[idx] = sound
return text
def add_punctuation(text):
if len(text) < 1:
return text
if len(text) < 10:
if text[-1] in punctuations:
if text[-1] != ".":
return text[:-1] + "."
if text[-1] not in punctuations:
text += '.'
return text
def break_chunks(l, n):
"""Yield successive n-sized chunks from l."""
for i in range(0, len(l), n):
yield " ".join(l[i:i + n])
def split_by_threshold(text, threshold):
text_list = text.split()
if len(text_list) <= threshold:
return [text]
if threshold < len(text_list) < (threshold*2):
return list(break_chunks(
text_list,
int(math.ceil(len(text_list) / 2))
))
elif (threshold*2) < len(text_list) < (threshold*3):
return list(break_chunks(
text_list,
int(math.ceil(len(text_list) / 3))
))
elif (threshold*3) < len(text_list) < (threshold*4):
return list(break_chunks(
text_list,
int(math.ceil(len(text_list) / 4))
))
else:
return list(break_chunks(
text_list,
int(math.ceil(len(text_list) / 4))
))
def synthesize_helper(text, synthesizer, threshold=10):
text_list = text.split()
if len(text_list) <= threshold*1.3:
text = " ".join(replace_acronym(text_list))
print(text.encode('utf-8'))
wav, _ = synthesizer.synthesize(add_punctuation(text), return_wav=True)
out = io.BytesIO()
audio.save_wav(wav, out)
return out.getvalue()
split_by_punc = None
if len(text_list) >= threshold:
for punc in split_punctuations:
if punc in text:
split_by_punc = text.split(punc)
break
chunks = []
if split_by_punc:
for sentence in split_by_punc:
sentence = sentence.strip()
chunk = split_by_threshold(sentence, threshold)
chunks += split_by_threshold(sentence, threshold)
else:
chunks += split_by_threshold(text, threshold)
combined_wav = np.array([])
for idx, chunk in enumerate(chunks):
if len(chunk) > 0:
text = add_punctuation(chunk)
text = " ".join(replace_acronym(text.split()))
print(text.encode('utf-8'))
wav, _ = synthesizer.synthesize(text, return_wav=True)
combined_wav = np.concatenate((combined_wav, wav[:-880*6]))
out = io.BytesIO()
audio.save_wav(combined_wav, out)
return out.getvalue()

7
synthesizer.py
View File

@ -31,7 +31,7 @@ class Synthesizer:
saver = tf.train.Saver()
saver.restore(self.session, checkpoint_path)
def synthesize(self, text):
def synthesize(self, text, return_wav=False):
cleaner_names = [x.strip() for x in hparams.cleaners.split(',')]
seq = text_to_sequence(text, cleaner_names)
feed_dict = {
@ -51,6 +51,9 @@ class Synthesizer:
wav = wav[:audio_endpoint]
alignment = alignment[:, :alignment_endpoint]
if return_wav:
return wav, alignment
out = io.BytesIO()
audio.save_wav(wav, out)
return out.getvalue(), alignment
return out, alignment