add: audiocraft spectrogram visualization enhancements (#459)

* add: dynamic range and better cmap

* add: frequency axis scaling based on mean spectrum threshold

* chore: changed string quotes to "

colabs/audiocraft/AudioCraft_MusicGen.ipynb

@@ -4,12 +4,7 @@
       "cell_type": "code",
       "execution_count": null,
       "metadata": {
-        "colab": {
-          "base_uri": "https://localhost:8080/",
-          "height": 1000
-        },
-        "id": "EZU3hg4B1om6",
-        "outputId": "1512f321-698c-4d01-e8d9-6d9577eee7e1"
+        "id": "EZU3hg4B1om6"
       },
       "outputs": [],
       "source": [
@@ -44,12 +39,7 @@
       "cell_type": "code",
       "execution_count": null,
       "metadata": {
-        "colab": {
-          "base_uri": "https://localhost:8080/",
-          "height": 122
-        },
-        "id": "3MTX8GoE7AzN",
-        "outputId": "33e92860-9ff9-4923-87a5-66c2d89e5f22"
+        "id": "3MTX8GoE7AzN"
       },
       "outputs": [],
       "source": [
@@ -119,11 +109,7 @@
       "cell_type": "code",
       "execution_count": null,
       "metadata": {
-        "colab": {
-          "base_uri": "https://localhost:8080/"
-        },
-        "id": "SfM8rhVX7ES9",
-        "outputId": "22fd7b60-ce23-4aac-86cf-5d5e0a712815"
+        "id": "SfM8rhVX7ES9"
       },
       "outputs": [],
       "source": [
@@ -157,42 +143,40 @@
       "outputs": [],
       "source": [
         "def get_spectrogram(audio_file, output_file):\n",
-        "    # Read the audio file\n",
         "    sample_rate, samples = wavfile.read(audio_file)\n",
-        "\n",
-        "    # Compute the spectrogram\n",
         "    frequencies, times, Sxx = signal.spectrogram(samples, sample_rate)\n",
         "\n",
-        "    # Create a figure and axis for plotting\n",
-        "    fig, ax = plt.subplots()\n",
+        "    log_Sxx = 10 * np.log10(Sxx + 1e-10)\n",
+        "    vmin = np.percentile(log_Sxx, 5)\n",
+        "    vmax = np.percentile(log_Sxx, 95)\n",
         "\n",
-        "    # Plot the spectrogram\n",
-        "    ax.pcolormesh(times, frequencies, 10 * np.log10(Sxx), shading='gouraud')\n",
+        "    mean_spectrum = np.mean(log_Sxx, axis=1)\n",
+        "    threshold_low = np.percentile(mean_spectrum, 5)\n",
+        "    threshold_high = np.percentile(mean_spectrum, 95)\n",
         "\n",
-        "    # Remove axis, labels, and other decorations\n",
-        "    ax.axis('off')\n",
-        "    plt.subplots_adjust(left=0, right=1, top=1, bottom=0)\n",
+        "    freq_indices = np.where(mean_spectrum > threshold_low)\n",
+        "    freq_min = 20\n",
+        "    freq_max = frequencies[freq_indices].max()\n",
         "\n",
-        "    # Save the figure to a temporary buffer\n",
-        "    plt.savefig(output_file, format='png', bbox_inches='tight', pad_inches=0)\n",
+        "    fig, ax = plt.subplots()\n",
+        "    cmap = plt.get_cmap(\"magma\")\n",
+        "\n",
+        "    ax.pcolormesh(times, frequencies, log_Sxx, shading=\"gouraud\", cmap=cmap, vmin=vmin, vmax=vmax)\n",
+        "    ax.axis(\"off\")\n",
+        "    ax.set_ylim([freq_min, freq_max])\n",
         "\n",
-        "    # Close the plot\n",
+        "    plt.subplots_adjust(left=0, right=1, top=1, bottom=0)\n",
+        "    plt.savefig(output_file, format=\"png\", bbox_inches=\"tight\", pad_inches=0)\n",
         "    plt.close()\n",
         "\n",
-        "    # Return the image saved in the buffer\n",
         "    return wandb.Image(output_file)"
       ]
     },
     {
       "cell_type": "code",
       "execution_count": null,
       "metadata": {
-        "colab": {
-          "base_uri": "https://localhost:8080/",
-          "height": 140
-        },
-        "id": "4fcOpiYx7Fqf",
-        "outputId": "4282158f-1170-4a3b-e514-a4f3916e3a7e"
+        "id": "4fcOpiYx7Fqf"
       },
       "outputs": [],
       "source": [
@@ -253,4 +237,4 @@
   },
   "nbformat": 4,
   "nbformat_minor": 0
-}
+}