def run_tuner():
import aukit
from aukit.audio_tuner import tune_speed, tune_pitch
inpath = r"hello.wav"
aukit.anything2bytes(inpath)
aukit.anything2wav(inpath)
aukit.anything2bytesio(inpath)
bys = tune_speed(inpath, sr=16000, rate=0.5, out_type=None)
print(bys)
wav = tune_pitch(bys, sr=16000, rate=1, out_type=None)
print(wav)
aukit.play_audio(wav)
def tts_sdk_base(text, speaker='biaobei', audio='24', output='', **kwargs):
"""语音合成函数式SDK接口。
text为待合成的文本。
speaker可设置为内置的发音人名称，可选名称见_reference_audio_dict；默认的发音人名称列表见resource/reference_audio/__init__.py。
audio如果是数字，则调用内置的语音作为发音人参考音频；如果是语音路径，则调用audio路径的语音作为发音人参考音频。
output如果以.wav结尾，则为保存语音文件的路径；如果以play开头，则合成语音后自动播放语音。
"""
global _dataloader
if _dataloader is None:
load_models(**kwargs)
load_audio(**kwargs)
if str(audio).isdigit():
audio = _reference_audio_list[(int(audio) - 1) %
len(_reference_audio_list)]
elif os.path.isfile(audio):
audio = str(audio)
elif isinstance(audio, bytes):
tmp_audio = tempfile.TemporaryFile(suffix='.wav')
tmp_audio.write(audio)
audio = tmp_audio.name
elif isinstance(audio, str) and len(audio) >= 256:
tmp_audio = tempfile.TemporaryFile(suffix='.wav')
tmp_audio.write(base64.standard_b64decode(audio))
audio = tmp_audio.name
elif speaker in _reference_audio_dict:
audio = _reference_audio_dict[speaker]
else:
raise AssertionError
text_data, style_data, speaker_data, f0_data, mel_data = transform_mellotron_input_data(
dataloader=_dataloader,
text=text,
speaker=speaker,
audio=audio,
device=_device)
mels, mels_postnet, gates, alignments = mellotron.generate_mel(
text_data, style_data, speaker_data, f0_data)
out_gate = gates.cpu().numpy()[0]
end_idx = np.argmax(out_gate > kwargs.get('gate_threshold', 0.2)
) or np.argmax(out_gate) or out_gate.shape[0]
mels_postnet = mels_postnet[:, :, :end_idx]
vocoder_name = kwargs.get('vocoder', 'waveglow')
if vocoder_name == 'waveglow':
wavs = waveglow.generate_wave(mel=mels_postnet, **kwargs)
else:
wavs = _stft.griffin_lim(mels_postnet, n_iters=10)
wav_output = wavs.squeeze(0).cpu().numpy()
if output.startswith('play'):
aukit.play_sound(wav_output, sr=_stft.sampling_rate)
if output.endswith('.wav'):
aukit.save_wav(wav_output, output, sr=_stft.sampling_rate)
wav_output = aukit.anything2bytes(wav_output, sr=_stft.sampling_rate)
return wav_output
def tts_sdk(text, speaker='biaobei', audio='0', **kwargs):
global _dataloader
if _dataloader is None:
load_models(**kwargs)
if str(audio).isdigit():
audio = _reference_audio_list[(int(audio) - 1) %
len(_reference_audio_list)]
elif os.path.isfile(audio):
audio = str(audio)
elif isinstance(audio, bytes):
tmp_audio = tempfile.TemporaryFile(suffix='.wav')
tmp_audio.write(audio)
audio = tmp_audio.name
elif isinstance(audio, str) and len(audio) >= 100:
tmp_audio = tempfile.TemporaryFile(suffix='.wav')
tmp_audio.write(base64.standard_b64decode(audio))
audio = tmp_audio.name
elif speaker in _reference_audio_dict:
audio = _reference_audio_dict[speaker]
else:
raise AssertionError
text_data, style_data, speaker_data, f0_data, mel_data = transform_mellotron_input_data(
dataloader=_dataloader,
text=text,
speaker=speaker,
audio=audio,
device=_device)
mels, mels_postnet, gates, alignments = mellotron.generate_mel(
text_data, style_data, speaker_data, f0_data)
out_gate = gates.cpu().numpy()[0]
end_idx = np.argmax(
out_gate > 0.2) or np.argmax(out_gate) or out_gate.shape[0]
mels_postnet = mels_postnet[:, :, :end_idx]
if _use_waveglow:
wavs = waveglow.generate_wave(mel=mels_postnet, **kwargs)
else:
wavs = _stft.griffin_lim(mels_postnet, n_iters=5)
wav_output = wavs.squeeze(0).cpu().numpy()
output = kwargs.get('output', '')
if output.startswith('play'):
aukit.play_sound(wav_output, sr=_stft.sampling_rate)
if output.endswith('.wav'):
aukit.save_wav(wav_output, output, sr=_stft.sampling_rate)
wav_output = aukit.anything2bytes(wav_output, sr=_stft.sampling_rate)
return wav_output