Overview

Function Description
hpss(y, **kwargs) Decompose an audio time series into harmonic and percussive components.
harmonic(y, **kwargs) Extract harmonic elements from an audio time-series.
percussive(y, **kwargs) Extract percussive elements from an audio time-series.
time_stretch(y, *, rate, **kwargs) Time-stretch an audio series by a fixed rate.
pitch_shift(y, *, sr, n_steps[, …]) Shift the pitch of a waveform by n_steps steps.
remix(y, intervals, *[, align_zeros]) Remix an audio signal by re-ordering time intervals.
trim(y, *[, top_db, ref, frame_length, …]) Trim leading and trailing silence from an audio signal.
split(y, *[, top_db, ref, frame_length, …]) Split an audio signal into non-silent intervals.
preemphasis(y, *[, coef, zi, return_zf]) Pre-emphasize an audio signal with a first-order differencing filter.
deemphasis(y, *[, coef, zi, return_zf]) De-emphasize an audio signal with the inverse operation of preemphasis().

Harmonic and Percussive Source Separation

在音訊處理中，Harmonic-percussive source separation (HPSS) 是一種常用的技術，用於將音訊訊號區分為具有較強和較弱音高特徵的音訊元素，以及具有節奏和振幅變化的音訊元素。

  • hpss(y)：將聲音分成諧波和打擊部分
    • harmonic: 只回傳諧波部分
    • percussive: 只回傳打擊部分

Time and frequency

  • time_stretch(y, rate)：時間拉伸
    • rate：拉伸倍率
    • 不會改變聲音的頻率
  • pitch_shift(y, sr, n_steps)：音高變換
    • sr：取樣頻率
    • bins_per_octave：每個八度的半音數，預設為12
    • n_steps：音高變換的半音數

Miscellaneous

  • remix(y, intervals)：重新排列時間間隔
    • intervals：時間間隔
      • 例如： array([[ 1536, 10752],[10752, 20480],...])
  • trim(y, top_db)：去除開頭和結尾的靜音
    • top_db：靜音的閾值，預設為60
  • split(y, top_db)：將音訊分割為非靜音區間
    • top_db：靜音的閾值，預設為60
  • preemphasis(y, coef)：預強調
    • y[n]=x[n]−αx[n−1]y[n] = x[n] - \alpha x[n-1]
    • coef：預強調係數，預設為0.97
  • deemphasis(y, coef)：反預強調
    • y[n]=x[n]+αy[n−1]y[n] = x[n] + \alpha y[n-1]
    • coef：預強調係數，預設為0.97
    • 會和原本的可能有很小的不同，建議使用np.allclose(y, y_de, atol=1e-5)來測試

參考資料