audio_knowledge

记录音频相关知识

librosa相关

mel_spectrogram = librosa.feature.melspectrogram(y=wavsignal, sr=16000, n_mels=256, n_fft=1024, win_length=400, hop_length=160)

采样率是16000,表示1s内采样16000次,那么通常一帧为25ms,即一帧采样 16000 * 0.025 = 400 个点,所以 win_lehgth = 400,hop_lenght=160