gary on gary

app.py

@@ -11,6 +11,7 @@ from audiocraft.data.audio import audio_write
 from pydub import AudioSegment
 import spaces
 import tempfile
+from pydub import AudioSegment
 
 # Check if CUDA is available
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
@@ -198,7 +199,7 @@ def continue_music(input_audio_path, prompt_duration, musicgen_model, num_iterat
     # Prepare the audio slice for generation
     prompt_waveform = preprocess_audio(prompt_waveform)
     
-    # Load the model and set generation parameters as before
+    # Load the model and set generation parameters
     model_continue = MusicGen.get_pretrained(musicgen_model.split(" ")[0])
     model_continue.set_generation_params(
         use_sampling=True,
@@ -209,32 +210,39 @@ def continue_music(input_audio_path, prompt_duration, musicgen_model, num_iterat
         cfg_coef=3
     )
     
-    all_audio_files = []
+    original_audio = AudioSegment.from_mp3(input_audio_path)
+    all_audio_files = [original_audio]  # Start with the original audio
+    file_paths_for_cleanup = []  # List to track generated file paths for cleanup
+
     for i in range(num_iterations):
         output = model_continue.generate_continuation(prompt_waveform, prompt_sample_rate=sr, progress=True)
-        output = output.cpu()  # Ensure the output is on CPU for further processing
+        output = output.cpu()  # Move the output tensor back to CPU
         if len(output.size()) > 2:
             output = output.squeeze()
 
         filename_without_extension = f'continue_{i}'
         filename_with_extension = f'{filename_without_extension}.wav'
+        correct_filename_extension = f'{filename_without_extension}.wav.wav'  # Apply the workaround for audio_write
+
         audio_write(filename_with_extension, output, model_continue.sample_rate, strategy="loudness", loudness_compressor=True)
-        all_audio_files.append(filename_with_extension)
+        new_audio_segment = AudioSegment.from_wav(correct_filename_extension)
+        all_audio_files.append(new_audio_segment)
+        file_paths_for_cleanup.append(correct_filename_extension)  # Add to cleanup list
 
-    # Combine all audio files as before
-    combined_audio = AudioSegment.empty()
-    for filename in all_audio_files:
-        combined_audio += AudioSegment.from_wav(filename)
+    # Combine all audio files into one continuous segment
+    combined_audio = sum(all_audio_files)
 
     combined_audio_filename = f"combined_audio_{random.randint(1, 10000)}.mp3"
     combined_audio.export(combined_audio_filename, format="mp3")
 
-    # Clean up temporary files
-    for filename in all_audio_files:
-        os.remove(filename)
+    # Clean up temporary files using the list of file paths
+    for file_path in file_paths_for_cleanup:
+        os.remove(file_path)
 
     return combined_audio_filename
 
+
+
 # Define the expandable sections
 musiclang_blurb = """
 ## musiclang
@@ -289,9 +297,9 @@ with gr.Blocks() as iface:
             ], value="thepatch/vanya_ai_dnb_0.1 (small)")
             num_iterations = gr.Slider(label="Number of Iterations", minimum=1, maximum=3, step=1, value=3)
             generate_music_button = gr.Button("Generate Music")
-            output_audio = gr.Audio(label="Generated Music")
+            output_audio = gr.Audio(label="Generated Music", type="filepath")
             continue_button = gr.Button("Continue Generating Music")
-            continue_output_audio = gr.Audio(label="Continued Music Output")
+            continue_output_audio = gr.Audio(label="Continued Music Output", type="filepath")
 
     # Connecting the components
     generate_midi_button.click(generate_midi, inputs=[seed, use_chords, chord_progression, bpm], outputs=[midi_audio])