def synthesis_post():
global model, waveglow_model
if not model:
return redirect("/synthesis-setup")
if request.method == "GET":
return render_template("synthesis.html")
else:
text = request.form["text"]
folder_name = get_suffix()
results_folder = os.path.join(paths["results"], folder_name)
os.makedirs(results_folder)
graph_path = os.path.join(results_folder, GRAPH_FILE)
audio_path = os.path.join(results_folder, RESULTS_FILE)
graph_web_path = graph_path.replace("\\", "/")
audio_web_path = audio_path.replace("\\", "/")
synthesize(model, waveglow_model, text, inflect_engine, graph_path,
audio_path)
return render_template(
"synthesis.html",
text=text.strip(),
alertnative_words=get_alternative_word_suggestions(
audio_path, text),
graph=graph_web_path,
audio=audio_web_path,
)
def test_synthesize():
model_path = os.path.join("files", "tacotron2_statedict.pt")
waveglow_path = os.path.join("files",
"waveglow_256channels_universal_v5.pt")
graph_path = "graph.png"
audio_path = "synthesized_audio.wav"
model = load_model(model_path)
assert model
waveglow = load_waveglow_model(waveglow_path)
assert waveglow
text = "hello everybody my name is david attenborough"
inflect_engine = inflect.engine()
synthesize(model,
waveglow,
text,
inflect_engine,
graph=graph_path,
audio=audio_path)
assert text_similarity(text, transcribe(audio_path)) > 0.5
assert os.path.isfile(graph_path)
assert os.path.isfile(audio_path)
os.remove(graph_path)
os.remove(audio_path)
def synthesis_post():
global model, vocoder, symbols
if not model or not vocoder or not symbols:
return redirect("/synthesis-setup")
if request.method == "GET":
return render_template("synthesis.html")
else:
text = request.form.getlist("text")
if len(text) == 1:
text = text[0]
method = request.form["text_method"]
split_text = method == "paragraph"
parent_folder = os.path.join(paths["results"],
datetime.now().strftime("%Y-%m"))
os.makedirs(parent_folder, exist_ok=True)
first_line = text[0] if type(text) == list else text
results_folder = os.path.join(
parent_folder,
get_suffix() + "-" +
re.sub("[^0-9a-zA-Z _]+", "", first_line.replace(" ", "_"))[:20])
os.makedirs(results_folder)
graph_path = os.path.join(results_folder, GRAPH_FILE)
audio_path = os.path.join(results_folder, RESULTS_FILE)
graph_web_path = os.path.relpath(graph_path).replace("\\", "/")
audio_web_path = os.path.relpath(audio_path).replace("\\", "/")
silence = float(request.form["silence"])
max_decoder_steps = int(request.form["max_decoder_steps"])
synthesize(
model,
text,
symbols,
graph_path,
audio_path,
vocoder,
silence,
max_decoder_steps=max_decoder_steps,
split_text=split_text,
)
return render_template(
"synthesis.html",
text=text,
method=method,
graph=graph_web_path,
audio=audio_web_path,
silence=silence,
max_decoder_steps=max_decoder_steps,
)
def test_hifigan_synthesis():
hifigan_model_path = os.path.join("test_samples", "hifigan.pt")
hifigan_config_path = os.path.join("test_samples", "config.json")
audio_path = "synthesized_audio.wav"
transcription_model = Silero()
hifigan = Hifigan(hifigan_model_path, hifigan_config_path)
text = "the monkeys live"
synthesize(
model=FakeModelForSynthesis(),
text=text,
graph_path=None,
audio_path=audio_path,
vocoder=hifigan,
)
assert os.path.isfile(audio_path)
assert similarity(
text, transcription_model.transcribe(audio_path)) > MIN_SYNTHESIS_SCORE
os.remove(audio_path)
def test_synthesize():
graph_path = "graph.png"
audio_path = "synthesized_audio.wav"
model = FakeModel()
vocoder = FakeVocoder()
# Single line
text = "hello everybody my name is david attenborough"
synthesize(
model=model,
text=text,
graph_path=graph_path,
audio_path=audio_path,
vocoder=vocoder,
sample_rate=22050,
)
assert os.path.isfile(graph_path)
assert os.path.isfile(audio_path)
assert librosa.get_duration(filename=audio_path) == 1
os.remove(graph_path)
os.remove(audio_path)
# Multi line
text = [
"the monkeys live in the jungle with their families.",
"however, i prefer to live on the beach and enjoy the sun.",
]
synthesize(
model=model,
text=text,
graph_path=graph_path,
audio_path=audio_path,
vocoder=vocoder,
silence_padding=0.5,
sample_rate=22050,
)
assert os.path.isfile(graph_path)
assert os.path.isfile(audio_path)
assert librosa.get_duration(filename=audio_path) == 2.5
os.remove(graph_path)
os.remove(audio_path)
# Split text
text = (
"the monkeys live in the jungle with their families. however, i prefer to live on the beach and enjoy the sun."
)
synthesize(
model=model,
text=text,
graph_path=None,
audio_path=audio_path,
vocoder=vocoder,
silence_padding=0.5,
sample_rate=22050,
split_text=True,
)
assert os.path.isfile(audio_path)
assert librosa.get_duration(filename=audio_path) == 2.5
os.remove(audio_path)