Added Mel spectrogram (#1)

- Added Mel spectrogram (ee40ca2679686ea68ecdb4546b7d2fdac3a8c0ae)


Co-authored-by: Armin <[email protected]>

app.py

@@ -1,10 +1,11 @@
 import gradio as gr
 import matplotlib.pyplot as plt
-from matplotlib.colors import LinearSegmentedColormap
 import numpy as np
 import os
 import soundfile as sf
 import requests
+import librosa.display
+
 
 def download_file(url):
     file_id = url.split('/')[-2]
@@ -14,16 +15,23 @@ def download_file(url):
     open(local_filename, 'wb').write(response.content)
     return local_filename
 
+
 def main():
     with gr.Blocks() as app:
         gr.Markdown(
             """
-            Audio Analyzer Software by Ilaria, Help me on Ko-Fi!\n
-            Special thanks to Alex Murkoff for helping me coding it!
-            Need help with AI? Join Join AI Hub!
+            <h1><center>Audio Analyzer by Ilaria</center></h1>\n
+            <h3><center>Help me on <a href="https://ko-fi.com/ilariaowo/shop">Ko-Fi</a>!</center></h3>\n
+            ## Special thanks to Alex Murkoff for helping me code it!
+            #### Need help with AI? Join [AI Hub](https://discord.gg/aihub)!\n
+            **Note**: Try to keep the audio length under **2 minutes**,
+            since long audio files dont work well with a static spectrogram
             """
         )
-    
+
+        with gr.Row():
+            image_output = gr.Image(type='filepath', interactive=False)
+
         with gr.Row():
             with gr.Column():
                 audio_input = gr.Audio(type='filepath')
@@ -31,56 +39,44 @@ def main():
 
             with gr.Column():
                 output_markdown = gr.Markdown(value="", visible=True)
-                image_output = gr.Image(type='filepath', interactive=False)
-            
+
                 with gr.Accordion('Audio Downloader', open=False):
                     url_input = gr.Textbox(value='', label='Google Drive Audio URL')
                     download_butt = gr.Button(value='Download audio', variant='primary')
-                
+
                 download_butt.click(fn=download_file, inputs=[url_input], outputs=[audio_input])
-                create_spec_butt.click(fn=create_spectrogram_and_get_info, inputs=[audio_input], outputs=[output_markdown, image_output])
-            
+                create_spec_butt.click(fn=create_spectrogram_and_get_info, inputs=[audio_input],
+                                       outputs=[output_markdown, image_output])
+
         download_butt.click(fn=download_file, inputs=[url_input], outputs=[audio_input])
-        create_spec_butt.click(fn=create_spectrogram_and_get_info, inputs=[audio_input], outputs=[output_markdown, image_output])
-        
-        app.queue(max_size=1022).launch(share=True)
+        create_spec_butt.click(fn=create_spectrogram_and_get_info, inputs=[audio_input],
+                               outputs=[output_markdown, image_output])
 
-from matplotlib.colors import LinearSegmentedColormap
+        app.queue(max_size=1022).launch(share=True)
 
 def create_spectrogram_and_get_info(audio_file):
     plt.clf()
-    
-    cdict = {'red':   [(0.0,  0.0, 0.0),
-                       (0.5,  1.0, 1.0),
-                       (1.0,  1.0, 1.0)],
-             'green': [(0.0,  0.0, 0.0),
-                       (0.25, 0.0, 0.0),
-                       (0.75, 1.0, 1.0),
-                       (1.0,  1.0, 1.0)],
-             'blue':  [(0.0,  0.0, 0.0),
-                       (0.5,  0.0, 0.0),
-                       (1.0,  0.0, 0.0)]}
-    custom_cmap = LinearSegmentedColormap('CustomMap', cdict)
-    
-    fig = plt.figure(figsize=(15, 5))
-    fig.patch.set_facecolor('black')  # Imposta il colore di sfondo su nero
-    audio_data, sample_rate = sf.read(audio_file)
-    if len(audio_data.shape) > 1:
-        audio_data = np.mean(audio_data, axis=1)
-    plt.specgram(audio_data, Fs=sample_rate / 1, NFFT=4096, sides='onesided',
-                 cmap=custom_cmap, scale_by_freq=True, scale='dB', mode='magnitude', window=np.hanning(4096))  # Usa la mappa di colori personalizzata
-    plt.savefig('spectrogram.png', dpi=300)
+
+    y, sr = librosa.load(audio_file, sr=None)
+    S = librosa.feature.melspectrogram(y, sr=sr, n_mels=256)
+    log_S = librosa.amplitude_to_db(S, ref=np.max, top_db=256)
+    plt.figure(figsize=(12, 5.5))
+    librosa.display.specshow(log_S, sr=sr, x_axis='time')
+    plt.colorbar(format='%+2.0f dB', pad=0.01)
+    plt.tight_layout(pad=0.5)
+    plt.savefig('spectrogram.png', dpi=500)
     audio_info = sf.info(audio_file)
     bit_depth = {'PCM_16': 16, 'FLOAT': 32}.get(audio_info.subtype, 0)
     minutes, seconds = divmod(audio_info.duration, 60)
     seconds, milliseconds = divmod(seconds, 1)
     milliseconds *= 1000
-    bitrate = audio_info.samplerate * audio_info.channels * bit_depth / 8 / 1024 / 1024
+    # bitrate = audio_info.samplerate * audio_info.channels * bit_depth / 8 / 1024 / 1024
+    # this bitrate one doesnt seem to be used anywhere so i just removed it
     speed_in_kbps = audio_info.samplerate * bit_depth / 1000
     filename_without_extension, _ = os.path.splitext(os.path.basename(audio_file))
     info_table = f"""
 
-    
+
     | Information | Value |
     | :---: | :---: |
     | File Name | {filename_without_extension} |
@@ -89,9 +85,9 @@ def create_spectrogram_and_get_info(audio_file):
     | Audio Channels | {audio_info.channels} |
     | Samples per second | {audio_info.samplerate} Hz |
     | Bit per second | {audio_info.samplerate * audio_info.channels * bit_depth} bit/s |
-    
+
     """
-    
+
     # Return the PNG file of the spectrogram and the info table
     return info_table, 'spectrogram.png'