Test program for librosa CQT algorithm

test_cqt.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Test program for librosa CQT algorithm.

https://gist.github.com/trueroad/c5861963278d97bd73e032f5afaad987

Copyright (C) 2023 Masamichi Hosoda.
All rights reserved.

Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions
are met:

* Redistributions of source code must retain the above copyright notice,
  this list of conditions and the following disclaimer.

* Redistributions in binary form must reproduce the above copyright notice,
  this list of conditions and the following disclaimer in the documentation
  and/or other materials provided with the distribution.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
ARE DISCLAIMED.
IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
SUCH DAMAGE.
"""

# The NPZ saved by this tool is similar in format to the NPZ generated by
#   Experiment chromatic DFT algorithm (chromatic_dft.py and test_cdft.py)
#   https://gist.github.com/trueroad/296dd6f397eb1ed65c1b04bc6a68c201
# and can be used by
#   Experiment frequency component vectors DTW (fcvs_dtw.py)
#   https://gist.github.com/trueroad/a5fee9ee0e447b78761dfdb336bbed7e
# and
#   Frequency component vectors and SMF (fcvs_smf.py)
#   https://gist.github.com/trueroad/216cb39902bccd776d8e9da670460fa3

# See
# https://www.wizard-notes.com/entry/python/librosa-cqt
# http://makotomurakami.com/blog/2020/06/08/5633/
# https://heartstat.net/2021/05/14/python_constant-q-transform/

import argparse
import os
from pathlib import Path
import sys
from typing import Any, Final, Union

import numpy as np
import numpy.typing as npt
import librosa


def generate_npz_by_cqt(filename_wav: Union[str, bytes, os.PathLike[Any]],
                        filename_npz: Union[str, os.PathLike[Any]],
                        shift_time: float = 0.0,
                        b_trim_leading: bool = False,
                        b_trim_trailing: bool = False,
                        top_db: int = 25,
                        sliding_size: float = 0.05,
                        filter_scale: float = 2.0
                        ) -> None:
    """
    Generate NPZ by CQT.

    Args:
      filename_wav (PathLike): Input WAV filename
      filename_npz (PathLike): Output NPZ filename
      shift_time (float): Shift time (positive: add beginning silence,
        negative: cut beginning waveform, default=0.0, i.e. 0.0 [s])
      b_trim_leading (bool): Trim leading silence
      b_trim_trailing (bool): Trim trailing silence
      trim_top_db (int): Trim's threshold in dB
        The smaller the value, the longer the length to trim
        and the shorter the length to remain.
      sliding_size (float): Sliding size (CQT hop length) in second
        (default=0.05, i.e. 50 [ms])
      filter_scale (float): CQT filter scale
    """
    fmin_str: Final[str] = 'C1'
    octaves: Final[int] = 8
    bins_per_octave: Final[int] = 12
    # window: Final[str] = 'hann'
    window: Final[str] = 'blackman'

    data: npt.NDArray[Any]
    sampling_rate: float
    data, sampling_rate = librosa.load(filename_wav, sr=None, mono=True)

    index: npt.NDArray[Any]
    if b_trim_leading or b_trim_trailing:
        _, index = librosa.effects.trim(data, top_db=top_db)
    if b_trim_leading and b_trim_trailing:
        data = data[index[0]:index[1]]
    elif b_trim_leading:
        data = data[index[0]:]
    elif b_trim_trailing:
        data = data[:index[1]]

    if shift_time > 0.0:
        silence: npt.NDArray[Any] \
            = np.zeros(int(shift_time * sampling_rate), dtype=data.dtype)
        data = np.concatenate([silence, data])
    elif shift_time < 0.0:
        data = data[int(-shift_time * sampling_rate):]

    cqt: npt.NDArray[np.complex64]
    cqt = librosa.cqt(data,
                      sr=sampling_rate,
                      hop_length=int(sampling_rate * sliding_size),
                      fmin=librosa.note_to_hz(fmin_str),
                      n_bins=octaves * bins_per_octave,
                      bins_per_octave=bins_per_octave,
                      filter_scale=filter_scale,
                      window=window)

    power: npt.NDArray[np.float64]
    power = np.abs(cqt, dtype=np.float64) ** 2.0

    vectors: npt.NDArray[np.float64] = np.fliplr(np.rot90(power, -1))

    np.savez_compressed(filename_npz, chroma_vectors=vectors)


def main() -> None:
    """Test main."""
    parser: argparse.ArgumentParser = argparse.ArgumentParser()
    parser.add_argument('DIR', help='Input WAV/Output NPZ directory '
                        '(recursive)')
    parser.add_argument('--trim-leading',
                        help='Trim leading silence',
                        action='store_true')
    parser.add_argument('--trim-trailing',
                        help='Trim trailing silence',
                        action='store_true')
    parser.add_argument('--trim-top-db',
                        help="Trim's threshold in dB "
                        '(default=25)',
                        type=int, default=25)
    parser.add_argument('--shift-time',
                        help='Shift time (positive: add beginning silence, '
                        'negative: cut beginning waveform, '
                        'default=0.0, i.e. 0.0 [s])',
                        type=float, default=0.0)
    parser.add_argument('--sliding-size',
                        help='Sliding size (CQT hop length) in second '
                        '(default=0.05, i.e. 50 [ms])',
                        type=float, default=0.05)
    parser.add_argument('--filter-scale',
                        help='CQT filter scale '
                        '(default=2.0)',
                        type=float, default=2.0)
    args = parser.parse_args()
    vargs = vars(args)

    dirname: str = vargs['DIR']
    b_trim_leading: bool = vargs['trim_leading']
    b_trim_trailing: bool = vargs['trim_trailing']
    trim_top_db: int = args.trim_top_db
    shift_time: float = args.shift_time
    sliding_size_time: float = args.sliding_size
    filter_scale: float = args.filter_scale

    print(f'Directory    : {dirname}\n'
          f'Trim leading : {b_trim_leading}\n'
          f'Trim trailing: {b_trim_trailing}\n'
          f'Trim top db  : {trim_top_db} [dB]\n'
          f'Shift time   : {shift_time} [s]\n'
          f'Sliding Size : {sliding_size_time} [s]\n'
          f'Filter scale : {filter_scale}\n')

    path_wav: Path
    for path_wav in Path(dirname).glob('**/*.wav'):
        path_npz: Path = path_wav.with_suffix('.npz')
        print(f'{path_wav}\n'
              f' -> {path_npz}')

        generate_npz_by_cqt(filename_wav=path_wav,
                            filename_npz=path_npz,
                            b_trim_leading=b_trim_leading,
                            b_trim_trailing=b_trim_trailing,
                            shift_time=shift_time,
                            sliding_size=sliding_size_time,
                            filter_scale=filter_scale)


if __name__ == '__main__':
    main()

test_npz2png.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Test NPZ to PNG.

https://gist.github.com/trueroad/c5861963278d97bd73e032f5afaad987

Copyright (C) 2023 Masamichi Hosoda.
All rights reserved.

Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions
are met:

* Redistributions of source code must retain the above copyright notice,
  this list of conditions and the following disclaimer.

* Redistributions in binary form must reproduce the above copyright notice,
  this list of conditions and the following disclaimer in the documentation
  and/or other materials provided with the distribution.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
ARE DISCLAIMED.
IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
SUCH DAMAGE.
"""

# This program generates a PNG from the NPZ generaged by test_cqt.py.
# The PNG is similar to the one generated by
#   Experiment chromatic DFT algorithm (chromatic_dft.py and test_cdft.py)
#   https://gist.github.com/trueroad/296dd6f397eb1ed65c1b04bc6a68c201

import argparse
import sys
from typing import Any

from PIL import Image
import numpy as np
import numpy.typing as npt


def main() -> None:
    """Test main."""
    parser: argparse.ArgumentParser = argparse.ArgumentParser()
    parser.add_argument('INPUT.npz', help='Input NPZ filename')
    parser.add_argument('OUTPUT.png', help='Output PNG filename')
    parser.add_argument('--gamma',
                        help='Encoding gamma for PNG (default=0.45)',
                        type=float, default=0.45)
    parser.add_argument('--vscale',
                        help='V scaling factor for PNG (default=4)',
                        type=int, default=4)
    parser.add_argument('--hscale',
                        help='H scaling factor for PNG (default=1)',
                        type=int, default=1)
    args = parser.parse_args()
    vargs = vars(args)

    filename_npz: str = vargs['INPUT.npz']
    filename_png: str = vargs['OUTPUT.png']
    gamma: float = args.gamma
    vscale: int = args.vscale
    hscale: int = args.hscale

    print(f'Input NPZ filename : {filename_npz}\n'
          f'Output PNG filename: {filename_png}\n'
          f'Encoding gamma     : {gamma}\n'
          f'V scale            : {vscale}\n'
          f'H scale            : {hscale}\n')

    loaded: np.lib.npyio.NpzFile = np.load(filename_npz)

    vectors: npt.NDArray[np.float64] = loaded['chroma_vectors']

    cvs_uint: npt.NDArray[np.uint8] = \
        ((vectors / vectors.max()) ** gamma
         * 255).astype(np.uint8)

    pil_img = Image.fromarray(np.rot90(cvs_uint))
    x, y = pil_img.size
    pil_img.resize(size=(x * hscale, y * vscale),
                   resample=Image.NEAREST).save(filename_png)


if __name__ == '__main__':
    main()