Spectrograms/python/examples/stmtf.py at main · jmg049/Spectrograms · GitHub

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
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
"""
Spectro-Temporal Modulation Transfer Function (STMTF)

Expected peaks:
    Temporal modulation  ≈ ±6 Hz
    Spectral modulation  ≈ ±0.125 cycles / bin
"""

import numpy as np
import spectrograms as sp
import matplotlib.pyplot as plt


def main():
    # -----------------------------
    # Signal parameters
    # -----------------------------
    sample_rate = 16_000.0
    duration = 3.0
    n_samples = int(sample_rate * duration)
    t = np.arange(n_samples) / sample_rate

    # -----------------------------
    # Travelling ripple parameters
    # -----------------------------
    fm = 6.0  # temporal modulation (Hz)
    cycles_across_band = 8.0  # spectral cycles across tone bank
    alpha = 0.9  # modulation depth

    n_tones = 64
    f_min, f_max = 300.0, 5_000.0
    freqs = np.geomspace(f_min, f_max, n_tones)

    print("Injected modulation:")
    print(f"  Temporal:  ±{fm:.2f} Hz")
    print(f"  Spectral:  ±{cycles_across_band / n_tones:.3f} cycles/bin\n")

    # -----------------------------
    # Generate signal
    # -----------------------------
    signal = np.zeros_like(t)

    for k, fk in enumerate(freqs):
        phase_ramp = 2 * np.pi * cycles_across_band * (k / n_tones)
        env = 1.0 + alpha * np.cos(2 * np.pi * fm * t + phase_ramp)
        signal += env * np.sin(2 * np.pi * fk * t)

    signal /= np.max(np.abs(signal)) + 1e-12

    # -----------------------------
    # Linear spectrogram (NOT mel)
    # -----------------------------
    stft = sp.StftParams(
        n_fft=512,
        hop_size=128,
        window=sp.WindowType.hanning,
        centre=True,
    )
    params = sp.SpectrogramParams(stft=stft, sample_rate=sample_rate)

    print("Computing linear spectrogram...")
    spectrogram = sp.compute_linear_power_spectrogram(signal, params)

    print("Spectrogram shape:", spectrogram.shape)
    print("Duration:", spectrogram.duration(), "s\n")

    # -----------------------------
    # Remove DC + normalise
    # -----------------------------
    spec = np.ascontiguousarray(spectrogram.T)
    spec -= spec.mean()
    spec /= spec.std() + 1e-12
    spec -= spec.mean(axis=1, keepdims=True)  # remove per-frequency DC
    spec -= spec.mean(axis=0, keepdims=True)  # remove per-time DC
    # -----------------------------
    # STMTF
    # -----------------------------
    print("Computing STMTF...")
    stmtf_mag = sp.magnitude_spectrum_2d(spec)
    stmtf = sp.fftshift(stmtf_mag)

    # -----------------------------
    # Modulation axes
    # -----------------------------
    n_freq_bins, n_time_frames = spec.shape

    hop = params.stft.hop_size
    frame_period = hop / sample_rate

    spectral_mod = sp.fftshift_1d(
        sp.fftfreq(n_freq_bins, d=1.0)  # cycles per bin
    )
    temporal_mod = sp.fftshift_1d(
        sp.fftfreq(n_time_frames, d=frame_period)  # Hz
    )

    # -----------------------------
    # Locate strongest non-DC peak
    # -----------------------------
    h, w = stmtf.shape
    mask = np.ones_like(stmtf, dtype=bool)
    mask[h // 2 - 2 : h // 2 + 3, w // 2 - 2 : w // 2 + 3] = False

    flat_idx = np.argmax(stmtf[mask])
    coords = np.argwhere(mask)[flat_idx]
    i, j = coords

    print("Measured peak:")
    print(f"  Spectral modulation: {spectral_mod[i]:.4f} cycles/bin")
    print(f"  Temporal modulation: {temporal_mod[j]:.4f} Hz\n")

    # -----------------------------
    # Visualisation
    # -----------------------------
    fig, axes = plt.subplots(1, 2, figsize=(14, 5))

    # Spectrogram
    im1 = axes[0].imshow(
        spec,
        aspect="auto",
        origin="lower",
        cmap="viridis",
    )
    axes[0].set_title("Linear Spectrogram (normalised)")
    axes[0].set_xlabel("Frame")
    axes[0].set_ylabel("Frequency bin")
    plt.colorbar(im1, ax=axes[0])

    # STMTF (log + clipped dynamic range)
    img = np.log10(stmtf + 1e-12)
    vmax = np.percentile(img, 99.5)
    vmin = vmax - 6.0

    tmin, tmax = temporal_mod.min(), temporal_mod.max()
    smin, smax = spectral_mod.min(), spectral_mod.max()

    im2 = axes[1].imshow(
        img,
        aspect="auto",
        origin="lower",
        cmap="hot",
        vmin=vmin,
        vmax=vmax,
        extent=[tmin, tmax, smin, smax],
    )

    axes[1].set_title("STMTF (log magnitude)")
    axes[1].set_xlabel("Temporal modulation (Hz)")
    axes[1].set_ylabel("Spectral modulation (cycles/bin)")
    plt.colorbar(im2, ax=axes[1])

    plt.tight_layout()
    plt.show()


if __name__ == "__main__":
    main()