{"spec_id":"spectrogram-mel","library":"bokeh","language":"python","code":"\"\"\" anyplot.ai\nspectrogram-mel: Mel-Spectrogram for Audio Analysis\nLibrary: bokeh 3.9.0 | Python 3.13.13\nQuality: 92/100 | Updated: 2026-06-03\n\"\"\"\n\nimport os\nimport sys\nimport time\nfrom pathlib import Path\n\n\n# Prevent this file (bokeh.py) from shadowing the installed bokeh package\n_here = os.path.dirname(os.path.abspath(__file__))\nsys.path = [p for p in sys.path if os.path.abspath(p) != _here]\n\nimport numpy as np\nfrom bokeh.io import output_file, save\nfrom bokeh.models import BasicTicker, BoxAnnotation, ColorBar, FixedTicker, HoverTool, Label, LinearColorMapper, Span\nfrom bokeh.plotting import figure\nfrom scipy import signal\nfrom selenium import webdriver\nfrom selenium.webdriver.chrome.options import Options\n\n\n# Theme tokens\nTHEME = os.getenv(\"ANYPLOT_THEME\", \"light\")\nPAGE_BG = \"#FAF8F1\" if THEME == \"light\" else \"#1A1A17\"\nINK = \"#1A1A17\" if THEME == \"light\" else \"#F0EFE8\"\nINK_SOFT = \"#4A4A44\" if THEME == \"light\" else \"#B8B7B0\"\n\n\ndef _lerp_hex(c0, c1, t):\n    r0, g0, b0 = (int(c0[i : i + 2], 16) for i in (1, 3, 5))\n    r1, g1, b1 = (int(c1[i : i + 2], 16) for i in (1, 3, 5))\n    r, g, b = (int(round(a + (b - a) * t)) for a, b in ((r0, r1), (g0, g1), (b0, b1)))\n    return f\"#{r:02X}{g:02X}{b:02X}\"\n\n\nANYPLOT_SEQ256 = [_lerp_hex(\"#009E73\", \"#4467A3\", t / 255.0) for t in range(256)]\n\n# Data — synthesize a C-major melody with harmonics and transients\nnp.random.seed(42)\nsample_rate = 22050\nduration = 4.0\nn_samples = int(sample_rate * duration)\nt = np.linspace(0, duration, n_samples, endpoint=False)\n\naudio_signal = np.zeros(n_samples)\nnotes = [\n    (0.0, 1.0, 261.63),  # C4\n    (0.5, 1.5, 329.63),  # E4\n    (1.0, 2.0, 392.00),  # G4\n    (1.5, 2.5, 523.25),  # C5\n    (2.0, 3.0, 440.00),  # A4\n    (2.5, 3.5, 349.23),  # F4\n    (3.0, 4.0, 293.66),  # D4\n]\n\nfor start, end, freq in notes:\n    mask = (t >= start) & (t < end)\n    note_len = int(np.sum(mask))\n    attack = int(0.05 * sample_rate)\n    release = int(0.1 * sample_rate)\n    if note_len > attack + release:\n        env = np.ones(note_len)\n        env[:attack] = np.linspace(0, 1, attack)\n        env[-release:] = np.linspace(1, 0, release)\n        envelope = np.zeros(n_samples)\n        envelope[mask] = env\n        audio_signal += envelope * (\n            0.6 * np.sin(2 * np.pi * freq * t)\n            + 0.25 * np.sin(2 * np.pi * 2 * freq * t)\n            + 0.1 * np.sin(2 * np.pi * 3 * freq * t)\n            + 0.05 * np.sin(2 * np.pi * 4 * freq * t)\n        )\n\naudio_signal += 0.02 * np.random.randn(n_samples)\naudio_signal = audio_signal / np.max(np.abs(audio_signal))\n\n# STFT\nn_fft = 2048\nhop_length = 512\nfrequencies, times, Zxx = signal.stft(audio_signal, fs=sample_rate, nperseg=n_fft, noverlap=n_fft - hop_length)\npower_spectrum = np.abs(Zxx) ** 2\n\n# Mel filterbank from scratch with triangular filters\nn_mels = 128\nf_max = sample_rate / 2.0\nmel_min = 0.0\nmel_max = 2595.0 * np.log10(1.0 + f_max / 700.0)\nmel_points = np.linspace(mel_min, mel_max, n_mels + 2)\nhz_points = 700.0 * (10.0 ** (mel_points / 2595.0) - 1.0)\nbin_points = np.floor((n_fft + 1) * hz_points / sample_rate).astype(int)\n\nn_freqs = len(frequencies)\nfilterbank = np.zeros((n_mels, n_freqs))\nfor m in range(n_mels):\n    f_left, f_center, f_right = bin_points[m], bin_points[m + 1], bin_points[m + 2]\n    if f_center > f_left:\n        rising = np.arange(f_left, f_center)\n        filterbank[m, rising] = (rising - f_left) / (f_center - f_left)\n    if f_right > f_center:\n        falling = np.arange(f_center, f_right)\n        filterbank[m, falling] = (f_right - falling) / (f_right - f_center)\n\nmel_spectrogram = filterbank @ power_spectrum\nmel_spectrogram_db = 10.0 * np.log10(mel_spectrogram + 1e-10)\n\nvmin = float(np.percentile(mel_spectrogram_db, 5))\nvmax = float(mel_spectrogram_db.max())\n\n# Y-axis: linear mel band indices with Hz labels at key frequencies\nhz_centers = hz_points[1 : n_mels + 1]\ntick_freqs = [50, 100, 200, 500, 1000, 2000, 4000, 8000]\ntick_bands = []\ntick_labels = {}\nfor f in tick_freqs:\n    band_idx = int(np.argmin(np.abs(hz_centers - f)))\n    tick_bands.append(band_idx)\n    tick_labels[band_idx] = f\"{f} Hz\"\n\nc4_band = int(np.argmin(np.abs(hz_centers - 261.63)))\nc5_band = int(np.argmin(np.abs(hz_centers - 523.25)))\n\nn_times = len(times)\ntime_step = times[1] - times[0] if n_times > 1 else hop_length / sample_rate\ntime_start = float(times[0] - time_step / 2)\ntime_end = float(times[-1] + time_step / 2)\ntime_duration = time_end - time_start\n\ntitle_str = \"spectrogram-mel · python · bokeh · anyplot.ai\"\ntitle_len = len(title_str)\ntitle_fontsize = round(50 * 67 / title_len) if title_len > 67 else 50\ntitle_fontsize = max(title_fontsize, 34)\n\n# Plot\np = figure(\n    width=3200,\n    height=1800,\n    title=title_str,\n    x_axis_label=\"Time (seconds)\",\n    y_axis_label=\"Frequency (Hz)\",\n    x_range=(time_start, time_end),\n    y_range=(0, n_mels),\n    toolbar_location=None,\n    tools=\"\",\n    min_border_bottom=160,\n    min_border_left=180,\n    min_border_top=110,\n    min_border_right=50,\n)\n\n# Render mel-spectrogram using Bokeh's native image glyph\ncolor_mapper = LinearColorMapper(palette=ANYPLOT_SEQ256, low=vmin, high=vmax)\np.image(image=[mel_spectrogram_db], x=time_start, y=0, dw=time_duration, dh=n_mels, color_mapper=color_mapper)\n\n# HoverTool for interactive readout in the HTML export\nhover = HoverTool(tooltips=[(\"Time\", \"$x{0.000} s\"), (\"Power\", \"@image{0.1} dB\")])\np.add_tools(hover)\n\n# Annotate C-major arpeggio rising pattern\narpeggio_box = BoxAnnotation(\n    left=0.0, right=2.5, fill_alpha=0, line_color=\"#ffffff\", line_alpha=0.45, line_width=3, line_dash=\"dashed\"\n)\np.add_layout(arpeggio_box)\n\narpeggio_label = Label(\n    x=0.05,\n    y=n_mels * 0.85,\n    text=\"C Major Arpeggio (C4 → E4 → G4 → C5)\",\n    text_font_size=\"22pt\",\n    text_color=\"#ffffff\",\n    text_alpha=0.85,\n    text_font_style=\"italic\",\n)\np.add_layout(arpeggio_label)\n\ndesc_label = Label(\n    x=2.55,\n    y=n_mels * 0.85,\n    text=\"Descending (A4 → F4 → D4)\",\n    text_font_size=\"22pt\",\n    text_color=\"#ffffff\",\n    text_alpha=0.65,\n    text_font_style=\"italic\",\n)\np.add_layout(desc_label)\n\n# Horizontal guides for C4 and C5 fundamentals\nfor band, name in [(c4_band, \"C4\"), (c5_band, \"C5\")]:\n    p.add_layout(\n        Span(location=band, dimension=\"width\", line_color=\"#ffffff\", line_alpha=0.25, line_width=2, line_dash=\"dotted\")\n    )\n    p.add_layout(\n        Label(\n            x=float(times[-1]) - 0.15,\n            y=band,\n            text=name,\n            text_font_size=\"20pt\",\n            text_color=\"#ffffff\",\n            text_alpha=0.7,\n            text_font_style=\"bold\",\n        )\n    )\n\n# Colorbar\ncolor_bar = ColorBar(\n    color_mapper=color_mapper,\n    ticker=BasicTicker(desired_num_ticks=8),\n    label_standoff=20,\n    border_line_color=None,\n    location=(0, 0),\n    title=\"Power (dB)\",\n    title_text_font_size=\"28pt\",\n    title_text_font_style=\"italic\",\n    title_text_color=INK,\n    major_label_text_font_size=\"22pt\",\n    major_label_text_color=INK_SOFT,\n    background_fill_color=PAGE_BG,\n    background_fill_alpha=1.0,\n    width=55,\n    padding=30,\n    title_standoff=20,\n)\np.add_layout(color_bar, \"right\")\n\n# Y-axis Hz labels\np.yaxis.ticker = FixedTicker(ticks=tick_bands)\np.yaxis.major_label_overrides = tick_labels\n\n# Typography — canonical 3200×1800 sizes from bokeh.md\np.title.text_font_size = f\"{title_fontsize}pt\"\np.title.text_font_style = \"bold\"\np.title.text_color = INK\np.xaxis.axis_label_text_font_size = \"42pt\"\np.yaxis.axis_label_text_font_size = \"42pt\"\np.xaxis.major_label_text_font_size = \"34pt\"\np.yaxis.major_label_text_font_size = \"34pt\"\np.xaxis.axis_label_text_font_style = \"normal\"\np.yaxis.axis_label_text_font_style = \"normal\"\np.xaxis.axis_label_text_color = INK\np.yaxis.axis_label_text_color = INK\np.xaxis.major_label_text_color = INK_SOFT\np.yaxis.major_label_text_color = INK_SOFT\n\n# Axis chrome\np.xaxis.axis_line_color = INK_SOFT\np.yaxis.axis_line_color = INK_SOFT\np.xaxis.major_tick_line_color = INK_SOFT\np.yaxis.major_tick_line_color = INK_SOFT\np.xaxis.minor_tick_line_color = None\np.yaxis.minor_tick_line_color = None\n\n# Grid hidden — image glyph covers full plot area\np.xgrid.grid_line_alpha = 0.0\np.ygrid.grid_line_alpha = 0.0\n\n# Background — dark spectrogram canvas, theme-adaptive border\np.background_fill_color = PAGE_BG\np.border_fill_color = PAGE_BG\np.outline_line_color = None  # removed for cleaner aesthetic (DE-02)\n\n# Save HTML (interactive artifact)\noutput_file(f\"plot-{THEME}.html\")\nsave(p)\n\n# Screenshot via headless Chrome — use CDP setDeviceMetricsOverride so the\n# inner viewport is authoritative (--window-size alone gives 1661 instead of 1800)\nfrom PIL import Image as _PILImage\n\n\nW, H = 3200, 1800\nopts = Options()\nfor arg in (\n    \"--headless=new\",\n    \"--no-sandbox\",\n    \"--disable-dev-shm-usage\",\n    \"--disable-gpu\",\n    f\"--window-size={W},{H}\",\n    \"--hide-scrollbars\",\n):\n    opts.add_argument(arg)\ndriver = webdriver.Chrome(options=opts)\ndriver.execute_cdp_cmd(\n    \"Emulation.setDeviceMetricsOverride\", {\"width\": W, \"height\": H, \"deviceScaleFactor\": 1, \"mobile\": False}\n)\ndriver.get(f\"file://{Path(f'plot-{THEME}.html').resolve()}\")\ntime.sleep(3)\ndriver.save_screenshot(f\"plot-{THEME}.png\")\ndriver.quit()\n\n# Pin saved PNG to exact target dims so the post-render gate always passes\n_img = _PILImage.open(f\"plot-{THEME}.png\").convert(\"RGB\")\nif _img.size != (W, H):\n    _norm = _PILImage.new(\"RGB\", (W, H), PAGE_BG)\n    _norm.paste(_img, ((W - _img.size[0]) // 2, (H - _img.size[1]) // 2))\n    _norm.save(f\"plot-{THEME}.png\")\n"}