DSP - Lab 2: 语谱图

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17

# main.py

def plot_waveform(filename: str, y: np.ndarray, sr: int) -> None:
    '''
    Plot the waveform of the audio signal.

    Args:
        `filename`: filename of the output figure
        `y`: time series of the audio signal
        `sr`: sample rate of the audio signal
    '''

    fig_time_path = fig_path / filename
    n_samples = y.shape[0]
    t = np.arange(n_samples) / sr
    utils.plot_time_domain(fig_time_path, t, y)
    print(f'Output figure to "{fig_time_path}".')

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26

# main.py

def create_spectrogram(y: np.ndarray, n_window: int) -> Tuple[np.ndarray, np.ndarray]:
    '''
    Create the spectrogram of the audio signal.

    Args:
        `y`: time series of the audio signal
        `n_window`: the number of samples used in each window

    Returns:
        `i_starts`: the starting indices of each window
        `spec`: the spectrum of frequencies of the audio signal as it varies with time
    '''

    n_samples = y.shape[0]
    i_starts = np.arange(0, n_samples, n_window // 2, dtype=int)
    i_starts = i_starts[i_starts + n_window < n_samples]
    n_fft = utils.next_pow2(n_window)
    zero_padding = np.zeros(n_fft - n_window)
    spec = np.array([np.abs(fft(
        np.concatenate((hamming(n_window) * y[i:i+n_window], zero_padding))
    )[:n_fft // 2]) for i in i_starts])
    # Rescale the absolute value of the spectrogram.
    spec = 10 * np.log10(spec.T + np.finfo(float).eps)
    return i_starts, spec

1
2
3

spec = np.array([np.abs(fft(
    np.concatenate((hamming(n_window) * y[i:i+n_window], zero_padding))
)[:n_fft // 2]) for i in i_starts])

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19

# windows.py

def hamming(m: int) -> np.ndarray:
    '''
    Return the Hamming window.

    Args:
        `m`: number of points in the output window

    Returns:
        The Hamming window of size `m`.
    '''

    if m < 1:
        return np.array([])
    if m == 1:
        return np.ones(1)
    n = np.arange(m)
    return 0.54 - 0.46 * np.cos(2 * np.pi * n / (m - 1))

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26

# main.py

fig_path = Path('assets/spectrogram/dev_set')

def plot_spectrogram(
    filename: str, i_starts: np.ndarray, spec: np.ndarray, sr: int
) -> None:
    '''
    Plot the spectrogram of the audio signal.

    Args:
        `filename`: filename of the output figure
        `i_starts`: the starting indices of each window
        `spec`: the spectrogram to plot
        `sr`: sample rate
    '''

    fig_spec_path = fig_path / filename
    xticks = np.linspace(0, spec.shape[1], 10)
    xlabels = [f'{i:4.2f}' for i in np.linspace(0, i_starts[-1] / sr, 10)]
    yticks = np.linspace(0, spec.shape[0], 10)
    ylabels = np.floor(fft_freq(spec.shape[0], sr, yticks)).astype(int)
    utils.plot_spectrogram(
        fig_spec_path, spec, xticks, xlabels, yticks, ylabels, n_window,
    )
    print(f'Output figure to "{fig_spec_path}".')

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34

# utils.py

def plot_spectrogram(
    output_path,
    spec: np.ndarray,
    xticks: np.ndarray,
    xlabels: np.ndarray,
    yticks: np.ndarray,
    ylabels: np.ndarray,
    n_fft: int,
) -> None:
    '''
    Plot the spectrogram of a wave.

    Args:
        `output_path`: path to the output figure
        `spec`: the spectrogram to plot
        `xticks`: tick locations of the x-axis
        `xlabels`: tick labels of the x-axis
        `yticks`: tick locations of the y-axis
        `ylabels`: tick labels of the y-axis
        `n_fft`: the number of samples for the FFT
    '''

    plt.figure()
    plt.title(f'Spectrogram ({n_fft} window size, hamming window)')
    plt.xticks(xticks, xlabels)
    plt.xlabel('Time / s')
    plt.yticks(yticks, ylabels)
    plt.ylabel('Frequency / Hz')
    plt.imshow(spec, origin='lower', aspect='auto')
    plt.colorbar(use_gridspec=True)
    plt.tight_layout()
    plt.savefig(output_path)

1
2

conda env update --name dsp --file environment.yml
conda activate dsp

1

python3 main.py

1
2

python3 resample.py "path/to/foobar.wav" 8000  # 单个文件
python3 resample.py "path/to/directory" 8000    # 目录下所有 .wav 文件递归批处理