alignment

alignment

pyampact.alignment.align_midi_wav(piece, WF, sr, TH, ST, width, tsr, nhar, wms, hop, showSpec)

Align a midi file to a wav file using the “peak structure distance” of Orio et al. that use the MIDI notes to build a mask that is compared against harmonics in the audio

Parameters:

• MF – Score instance of symbolic data

• WF – Audio time series of file

• TH – is the time step resolution (default 0.050)

• ST – is the similarity type: 0 (default) is triangle inequality

• showSpec – Boolean to show the spectrogram

Returns:

• m: Is the map s.t. M(:,m)

• [p,q]: Are the path from DP

• S: The similarity matrix

• D: Is the spectrogram

• M: Is the midi-note-derived mask

• N: Is Orio-style “peak structure distance”

pyampact.alignment.alignment_visualiser(audio_spec, times=None, freqs=None, fig=1, showSpec=True)

Plots a gross DTW alignment overlaid with the fine alignment resulting from the HMM aligner on the output of YIN. Trace(1,:) is the list of states in the HMM, and trace(2,:) is the number of YIN frames for which that state is occupied. Highlight is a list of notes for which the steady state will be highlighted.

Parameters:

• audio_spec – Spectrogram of audio file

• freqs – Array of sample frequencies

• times – Array of segment times

Returns:

Visualized spectrogram

pyampact.alignment.get_ons_offs(onsoffs)

Extracts a list of onset and offset from an inputted

3*N matrix of states and corresponding ending times from AMPACT’s HMM-based alignment algorithm

Parameters:

onsoffs – A 3*N alignment matrix, the first row is a list of N states the second row is the time which the state ends, and the third row is the state index

Returns:

• res.ons: List of onset times

• res.offs: List of offset times

pyampact.alignment.run_DTW_alignment(y, original_sr, piece, tres, width, target_sr, nharm, win_ms, hop, nmat, showSpec)

Perform a dynamic time warping alignment between specified audio and MIDI files.

Returns a matrix with the aligned onset and offset times (with corresponding MIDI note numbers) and a spectrogram of the audio.

Parameters:

• y – Audio time series of audio

• original_sr – original sample rate of audio

• piece – Score instance of symbolic data

• tres – Time resolution for MIDI to spectrum information conversion.

• width – Width parameter (you need to specify this value)

• target_sr – Target sample rate (you need to specify this value)

• nharm – Number of harmonics (you need to specify this value)

• win_ms – Window size in milliseconds (you need to specify this value)

• showSpec – Boolean to show the spectrogram

Returns:

• align: Dynamic time warping MIDI-audio alignment structure

  • align.on: onset times

  • align.off: offset times

  • align.midiNote: MIDI note numbers

• spec: Spectrogram of the audio file

• dtw: Dict of dynamic time warping returns

  • M: map s.t. M(:,m)

  • MA/RA [p,q]: path from DP

  • S: similarity matrix

  • D: spectrogram

  • notemask: midi-note-derived mask

  • pianoroll: midi-note-derived pianoroll

• nmat: updated DataFrame of nmat data

pyampact.alignment.run_alignment(y, original_sr, piece, nmat, width=3, target_sr=4000, nharm=3, win_ms=100, hop=32, showSpec=False)

Calls the DTW alignment function.

Parameters:

• y – Audio time series

• original_sr – original sample rate of audio

• piece – Score instance of symbolic data

• means – Mean values for each state

• covars – Covariance values for each state

• width – Width parameter (you need to specify this value)

• target_sr – Target sample rate (you need to specify this value)

• nharm – Number of harmonics (you need to specify this value)

• win_ms – Window size in milliseconds (you need to specify this value)

• hop – Number of samples between successive frames

• showSpec – Boolean to show the spectrogram

Returns:

• align: Dynamic time warping MIDI-audio alignment structure

  • on: onset times

  • off: offset times

  • midiNote: MIDI note numbers

• dtw: Dict of dynamic time warping returns

  • M: map s.t. M(:,m)

  • MA/RA [p,q]: path from DP

  • S: similarity matrix

  • D: spectrogram

  • notemask: midi-note-derived mask

  • pianoroll: midi-note-derived pianoroll

• spec: Spectrogram of the audio file

• newNmat: updated DataFrame of nmat data