def get_cqt()

in basic_pitch/models.py [0:0]

• 20 lines of code
• 2 McCabe index (conditional complexity)

def get_cqt(inputs: tf.Tensor, n_harmonics: int, use_batchnorm: bool) -> tf.Tensor:
    """Calculate the CQT of the input audio.

    Input shape: (batch, number of audio samples, 1)
    Output shape: (batch, number of frequency bins, number of time frames)

    Args:
        inputs: The audio input.
        n_harmonics: The number of harmonics to capture above the maximum output frequency.
            Used to calculate the number of semitones for the CQT.
        use_batchnorm: If True, applies batch normalization after computing the CQT

    Returns:
        The log-normalized CQT of the input audio.
    """
    n_semitones = np.min(
        [
            int(np.ceil(12.0 * np.log2(n_harmonics)) + ANNOTATIONS_N_SEMITONES),
            MAX_N_SEMITONES,
        ]
    )
    x = nn.FlattenAudioCh()(inputs)
    x = nnaudio.CQT(
        sr=AUDIO_SAMPLE_RATE,
        hop_length=FFT_HOP,
        fmin=ANNOTATIONS_BASE_FREQUENCY,
        n_bins=n_semitones * CONTOURS_BINS_PER_SEMITONE,
        bins_per_octave=12 * CONTOURS_BINS_PER_SEMITONE,
    )(x)
    x = signal.NormalizedLog()(x)
    x = tf.expand_dims(x, -1)
    if use_batchnorm:
        x = tfkl.BatchNormalization()(x)
    return x