def load_audio(audio_path, mode='train', win_length=400, sr=16000, hop_length=160, n_fft=512, spec_len=257):
# 读取音频数据
wav, sr_ret = librosa.load(audio_path, sr=sr)
# 推理的数据要移除静音部分
if mode == 'infer':
wav = remove_silence(wav, sr)
wav = remove_noise(wav, sr)
assert len(wav) > 0, "音频经过去除噪声和静音片段后，得到的结果为空，该音频不可用！"
# 数据拼接
if mode == 'train':
extended_wav = np.append(wav, wav)
if np.random.random() < 0.3:
extended_wav = extended_wav[::-1]
else:
extended_wav = np.append(wav, wav[::-1])
# 计算短时傅里叶变换
linear = librosa.stft(extended_wav, n_fft=n_fft, win_length=win_length, hop_length=hop_length)
linear_T = linear.T
mag, _ = librosa.magphase(linear_T)
mag_T = mag.T
freq, freq_time = mag_T.shape
assert freq_time >= spec_len, "非静音部分长度不能低于1.3s"
if mode == 'train':
# 随机裁剪
rand_time = np.random.randint(0, freq_time - spec_len)
spec_mag = mag_T[:, rand_time:rand_time + spec_len]
else:
spec_mag = mag_T[:, :spec_len]
mean = np.mean(spec_mag, 0, keepdims=True)
std = np.std(spec_mag, 0, keepdims=True)
spec_mag = (spec_mag - mean) / (std + 1e-5)
spec_mag = spec_mag[np.newaxis, :]
return spec_mag
def remove_noise_and_silence(wav, sr=_sr):
"""
remove noise and silence
:param wav:
:param sr:
:return:
"""
out = aukit.remove_noise(wav, sr=sr)
out = out / np.max(np.abs(out))
out = aukit.remove_silence_wave(out, sr=sr, keep_silence_len=50, min_silence_len=100, silence_thresh=-32)
out = aukit.strip_silence_wave(out, sr=sr, keep_silence_len=20, min_silence_len=40, silence_thresh=-32)
return out
def preprocess(self):
wav = self.ui.selected_utterance.wav
out = aukit.remove_noise(wav, sr=Synthesizer.sample_rate)
hp = aukit.Dict2Obj({})
hp["vad_window_length"] = 10 # milliseconds
hp["vad_moving_average_width"] = 2
hp["vad_max_silence_length"] = 2
hp["audio_norm_target_dBFS"] = -32
hp["sample_rate"] = 16000
hp["int16_max"] = (2**15) - 1
out = trim_long_silences(out, hparams=hp)
spec = Synthesizer.make_spectrogram(out)
self.ui.draw_align(spec[::-1], "current")
def run_noise_remover():
import aukit
inpath = r"hello.wav"
wav = aukit.load_wav(inpath)
out = aukit.remove_noise(wav)
aukit.play_audio(out)
def remove_noise_audio(inpath, outpath):
"""音频降噪。"""
import aukit
wav = aukit.load_wav(inpath, sr=16000)
out = aukit.remove_noise(wav, sr=16000)
aukit.save_wav(out, outpath, sr=16000)
def run_noise_remover():
import aukit
inpath = r"E:\data\temp\01.wav"
wav = aukit.load_wav(inpath)
out = aukit.remove_noise(wav)
aukit.play_audio(out)
