def synthesize(self):
self.ui.log("Generating the mel spectrogram...")
self.ui.set_loading(1)
# Synthesize the spectrogram
if self.synthesizer is None:
model_dir = self.ui.current_synthesizer_model_dir
checkpoints_dir = model_dir.joinpath("taco_pretrained")
self.synthesizer = Synthesizer(checkpoints_dir, low_mem=self.low_mem)
if not self.synthesizer.is_loaded():
self.ui.log("Loading the synthesizer %s" % self.synthesizer.checkpoint_fpath)
texts = self.ui.text_prompt.toPlainText().split("\n")
print(texts)
texts = g2p(texts)
print(texts)
embed = self.ui.selected_utterance.embed
embeds = np.stack([embed] * len(texts))
specs = self.synthesizer.synthesize_spectrograms(texts, embeds)
breaks = [spec.shape[1] for spec in specs]
spec = np.concatenate(specs, axis=1)
self.ui.draw_spec(spec, "generated")
self.current_generated = (self.ui.selected_utterance.speaker_name, spec, breaks, None)
self.ui.set_loading(0)
def preprocess_speaker_sst(speaker_dir, out_dir: Path, skip_existing: bool, hparams):
metadata = []
lines = []
texts = []
index = 1
with open(os.path.join(speaker_dir, 'metadata.csv'), encoding='utf-8') as f:
for line in f:
parts = line.strip().split(',')
#if (float(parts[2])>2 and float(parts[2])<7):
lines.append(parts[0])
with open(os.path.join(speaker_dir, parts[1]), encoding='utf-8') as f2:
for line2 in f2:
txt_paths = line2
texts.append(txt_paths)
#print(texts)
#print(lines)
texts = g2p(texts)
#print(texts)
#print(lines)
for basename, text in zip(lines,texts):
wav_path = os.path.join(speaker_dir, basename)
wav, _ = librosa.load(wav_path, hparams.sample_rate)
if hparams.rescale:
wav = wav / np.abs(wav).max() * hparams.rescaling_max
#futures.append(partial(process_utterance, mel_dir, linear_dir, wav_dir, basename, wav_path, text, hparams))
basename2 = basename.strip().split('/')
basename3 = "sl_"+basename2[0]+"_"+basename2[1]+"_"+basename2[2]
#+"_"+basename2[2]
#print(basename2[5])
metadata.append(process_utterance(wav, text, out_dir,basename3 ,
skip_existing, hparams))
index += 1
return [m for m in metadata if m is not None]
def preprocess_speaker2(speaker_dir, out_dir: Path, skip_existing: bool, hparams):
metadata = []
wavs = []
texts = []
for book_dir in speaker_dir.glob("*"):
# Gather the utterance audios and texts
try:
alignments_fpath = next(book_dir.glob("*.alignment.txt"))
with alignments_fpath.open("r") as alignments_file:
alignments = [line.rstrip().split(" ") for line in alignments_file]
except StopIteration:
# A few alignment files will be missing
continue
for wav_fname, words, end_times in alignments:
wav_fpath = book_dir.joinpath(wav_fname + ".flac")
assert wav_fpath.exists()
words = words.replace("\"", "").split(",")
end_times = list(map(float, end_times.replace("\"", "").split(",")))
# Process each sub-utterance
wav, text = split_on_silences(wav_fpath, words, end_times, hparams)
texts.extend(text)
wavs.extend(wav)
texts = g2p(texts)
for i, (wav, text) in enumerate(zip(wavs, texts)):
sub_basename = "%s_%02d" % (wav_fname, i)
metadata.append(process_utterance(wav, text, out_dir, sub_basename,
skip_existing, hparams))
return [m for m in metadata if m is not None]
def preprocess_speaker(speaker_dir, out_dir: Path, skip_existing: bool,
hparams):
metadata = []
lines = []
texts = []
index = 1
with open(os.path.join(speaker_dir, 'metadata.csv'),
encoding='utf-8') as f:
for line in f:
parts = line.strip().split('|')
basename = parts[0]
text = parts[2]
lines.append(basename)
texts.append(text)
texts = g2p(texts)
for basename, text in zip(lines, texts):
wav_path = os.path.join(speaker_dir, '{}.wav'.format(basename))
wav, _ = librosa.load(wav_path, hparams.sample_rate)
if hparams.rescale:
wav = wav / np.abs(wav).max() * hparams.rescaling_max
metadata.append(
process_utterance(wav, text, out_dir, basename, skip_existing,
hparams))
index += 1
return [m for m in metadata if m is not None]
preprocessed_wav = encoder.preprocess_wav(original_wav, sampling_rate)
print("Loaded file succesfully")
# Then we derive the embedding. There are many functions and parameters that the
# speaker encoder interfaces. These are mostly for in-depth research. You will typically
# only use this function (with its default parameters):
embed = encoder.embed_utterance(preprocessed_wav)
print("Created the embedding")
## Generating the spectrogram
# text = input("Write a sentence (+-20 words) to be synthesized:(Введите предложение для синтеза)\n")
# The synthesizer works in batch, so you need to put your data in a list or numpy array
texts = [args.text]
texts = g2p(texts)
print(texts)
embeds = [embed]
# If you know what the attention layer alignments are, you can retrieve them here by
# passing return_alignments=True
specs = synthesizer.synthesize_spectrograms(texts, embeds)
spec = specs[0]
print("Created the mel spectrogram")
## Generating the waveform
print("Synthesizing the waveform:")
# Synthesizing the waveform is fairly straightforward. Remember that the longer the
# spectrogram, the more time-efficient the vocoder.
generated_wav = vocoder.infer_waveform(spec)