ok herewego

app.py

@@ -18,9 +18,9 @@ def preprocess_audio(waveform):
     waveform_np = waveform.cpu().squeeze().numpy()
     return torch.from_numpy(waveform_np).unsqueeze(0).to(device)
 
-# Fix: Add dummy parameter to avoid schema generation bug
+# Test with a wrapper function
 @spaces.GPU
-def generate_drum_sample(dummy_trigger="generate"):
+def _generate_drum_sample_internal():
     model = MusicGen.get_pretrained('pharoAIsanders420/micro-musicgen-jungle')
     model.set_generation_params(duration=10)
     wav = model.generate_unconditional(1).squeeze(0)
@@ -32,140 +32,146 @@ def generate_drum_sample(dummy_trigger="generate"):
 
     return filename_with_extension
 
-@spaces.GPU
-def continue_drum_sample(existing_audio_path):
-    if existing_audio_path is None:
-        return None
+# Regular function wrapper (no @spaces.GPU on this one)
+def generate_drum_sample():
+    return _generate_drum_sample_internal()
+
+
+
+# @spaces.GPU
+# def continue_drum_sample(existing_audio_path):
+#     if existing_audio_path is None:
+#         return None
         
-    existing_audio, sr = torchaudio.load(existing_audio_path)
-    existing_audio = existing_audio.to(device)
+#     existing_audio, sr = torchaudio.load(existing_audio_path)
+#     existing_audio = existing_audio.to(device)
     
-    prompt_duration = 2
-    output_duration = 10
+#     prompt_duration = 2
+#     output_duration = 10
 
-    num_samples = int(prompt_duration * sr)
-    if existing_audio.shape[1] < num_samples:
-        raise ValueError("The existing audio is too short for the specified prompt duration.")
+#     num_samples = int(prompt_duration * sr)
+#     if existing_audio.shape[1] < num_samples:
+#         raise ValueError("The existing audio is too short for the specified prompt duration.")
 
-    start_sample = existing_audio.shape[1] - num_samples
-    prompt_waveform = existing_audio[..., start_sample:]
+#     start_sample = existing_audio.shape[1] - num_samples
+#     prompt_waveform = existing_audio[..., start_sample:]
 
-    model = MusicGen.get_pretrained('pharoAIsanders420/micro-musicgen-jungle')
-    model.set_generation_params(duration=output_duration)
+#     model = MusicGen.get_pretrained('pharoAIsanders420/micro-musicgen-jungle')
+#     model.set_generation_params(duration=output_duration)
 
-    output = model.generate_continuation(prompt_waveform, prompt_sample_rate=sr, progress=True)
-    output = output.to(device)
+#     output = model.generate_continuation(prompt_waveform, prompt_sample_rate=sr, progress=True)
+#     output = output.to(device)
 
-    if output.dim() == 3:
-        output = output.squeeze(0)
+#     if output.dim() == 3:
+#         output = output.squeeze(0)
 
-    if output.dim() == 1:
-        output = output.unsqueeze(0)
+#     if output.dim() == 1:
+#         output = output.unsqueeze(0)
 
-    combined_audio = torch.cat((existing_audio, output), dim=1)
-    combined_audio = combined_audio.cpu()
+#     combined_audio = torch.cat((existing_audio, output), dim=1)
+#     combined_audio = combined_audio.cpu()
 
-    combined_file_path = f'./continued_jungle_{random.randint(1000, 9999)}.wav'
-    torchaudio.save(combined_file_path, combined_audio, sr)
+#     combined_file_path = f'./continued_jungle_{random.randint(1000, 9999)}.wav'
+#     torchaudio.save(combined_file_path, combined_audio, sr)
 
-    return combined_file_path
+#     return combined_file_path
 
-@spaces.GPU
-def generate_music(wav_filename, prompt_duration, musicgen_model, output_duration):
-    if wav_filename is None:
-        return None
+# @spaces.GPU
+# def generate_music(wav_filename, prompt_duration, musicgen_model, output_duration):
+#     if wav_filename is None:
+#         return None
         
-    song, sr = torchaudio.load(wav_filename)
-    song = song.to(device)
-
-    model_name = musicgen_model.split(" ")[0]
-    model_continue = MusicGen.get_pretrained(model_name)
-
-    model_continue.set_generation_params(
-        use_sampling=True,
-        top_k=250,
-        top_p=0.0,
-        temperature=1.0,
-        duration=output_duration,
-        cfg_coef=3
-    )
-
-    prompt_waveform = song[..., :int(prompt_duration * sr)]
-    prompt_waveform = preprocess_audio(prompt_waveform)
+#     song, sr = torchaudio.load(wav_filename)
+#     song = song.to(device)
+
+#     model_name = musicgen_model.split(" ")[0]
+#     model_continue = MusicGen.get_pretrained(model_name)
+
+#     model_continue.set_generation_params(
+#         use_sampling=True,
+#         top_k=250,
+#         top_p=0.0,
+#         temperature=1.0,
+#         duration=output_duration,
+#         cfg_coef=3
+#     )
+
+#     prompt_waveform = song[..., :int(prompt_duration * sr)]
+#     prompt_waveform = preprocess_audio(prompt_waveform)
     
-    output = model_continue.generate_continuation(prompt_waveform, prompt_sample_rate=sr, progress=True)
-    output = output.cpu()
+#     output = model_continue.generate_continuation(prompt_waveform, prompt_sample_rate=sr, progress=True)
+#     output = output.cpu()
     
-    if len(output.size()) > 2:
-        output = output.squeeze()
+#     if len(output.size()) > 2:
+#         output = output.squeeze()
     
-    filename_without_extension = f'continued_music'
-    filename_with_extension = f'{filename_without_extension}.wav'
-    audio_write(filename_without_extension, output, model_continue.sample_rate, strategy="loudness", loudness_compressor=True)
+#     filename_without_extension = f'continued_music'
+#     filename_with_extension = f'{filename_without_extension}.wav'
+#     audio_write(filename_without_extension, output, model_continue.sample_rate, strategy="loudness", loudness_compressor=True)
 
-    return filename_with_extension
+#     return filename_with_extension
 
-@spaces.GPU
-def continue_music(input_audio_path, prompt_duration, musicgen_model, output_duration):
-    if input_audio_path is None:
-        return None
+# @spaces.GPU
+# def continue_music(input_audio_path, prompt_duration, musicgen_model, output_duration):
+#     if input_audio_path is None:
+#         return None
         
-    song, sr = torchaudio.load(input_audio_path)
-    song = song.to(device)
+#     song, sr = torchaudio.load(input_audio_path)
+#     song = song.to(device)
 
-    model_continue = MusicGen.get_pretrained(musicgen_model.split(" ")[0])
-    model_continue.set_generation_params(
-        use_sampling=True,
-        top_k=250,
-        top_p=0.0,
-        temperature=1.0,
-        duration=output_duration,
-        cfg_coef=3
-    )
+#     model_continue = MusicGen.get_pretrained(musicgen_model.split(" ")[0])
+#     model_continue.set_generation_params(
+#         use_sampling=True,
+#         top_k=250,
+#         top_p=0.0,
+#         temperature=1.0,
+#         duration=output_duration,
+#         cfg_coef=3
+#     )
 
-    original_audio = AudioSegment.from_mp3(input_audio_path)
-    current_audio = original_audio
+#     original_audio = AudioSegment.from_mp3(input_audio_path)
+#     current_audio = original_audio
 
-    file_paths_for_cleanup = []
+#     file_paths_for_cleanup = []
 
-    for i in range(1):
-        num_samples = int(prompt_duration * sr)
-        if current_audio.duration_seconds * 1000 < prompt_duration * 1000:
-            raise ValueError("The prompt_duration is longer than the current audio length.")
+#     for i in range(1):
+#         num_samples = int(prompt_duration * sr)
+#         if current_audio.duration_seconds * 1000 < prompt_duration * 1000:
+#             raise ValueError("The prompt_duration is longer than the current audio length.")
 
-        start_time = current_audio.duration_seconds * 1000 - prompt_duration * 1000
-        prompt_audio = current_audio[start_time:]
+#         start_time = current_audio.duration_seconds * 1000 - prompt_duration * 1000
+#         prompt_audio = current_audio[start_time:]
 
-        prompt_bytes = prompt_audio.export(format="wav").read()
-        prompt_waveform, _ = torchaudio.load(io.BytesIO(prompt_bytes))
-        prompt_waveform = prompt_waveform.to(device)
+#         prompt_bytes = prompt_audio.export(format="wav").read()
+#         prompt_waveform, _ = torchaudio.load(io.BytesIO(prompt_bytes))
+#         prompt_waveform = prompt_waveform.to(device)
 
-        prompt_waveform = preprocess_audio(prompt_waveform)
+#         prompt_waveform = preprocess_audio(prompt_waveform)
 
-        output = model_continue.generate_continuation(prompt_waveform, prompt_sample_rate=sr, progress=True)
-        output = output.cpu()
+#         output = model_continue.generate_continuation(prompt_waveform, prompt_sample_rate=sr, progress=True)
+#         output = output.cpu()
 
-        if len(output.size()) > 2:
-            output = output.squeeze()
+#         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'
+#         filename_without_extension = f'continue_{i}'
+#         filename_with_extension = f'{filename_without_extension}.wav'
+#         correct_filename_extension = f'{filename_without_extension}.wav.wav'
 
-        audio_write(filename_with_extension, output, model_continue.sample_rate, strategy="loudness", loudness_compressor=True)
-        generated_audio_segment = AudioSegment.from_wav(correct_filename_extension)
+#         audio_write(filename_with_extension, output, model_continue.sample_rate, strategy="loudness", loudness_compressor=True)
+#         generated_audio_segment = AudioSegment.from_wav(correct_filename_extension)
 
-        current_audio = current_audio[:start_time] + generated_audio_segment
+#         current_audio = current_audio[:start_time] + generated_audio_segment
 
-        file_paths_for_cleanup.append(correct_filename_extension)
+#         file_paths_for_cleanup.append(correct_filename_extension)
 
-    combined_audio_filename = f"combined_audio_{random.randint(1, 10000)}.mp3"
-    current_audio.export(combined_audio_filename, format="mp3")
+#     combined_audio_filename = f"combined_audio_{random.randint(1, 10000)}.mp3"
+#     current_audio.export(combined_audio_filename, format="mp3")
 
-    for file_path in file_paths_for_cleanup:
-        os.remove(file_path)
+#     for file_path in file_paths_for_cleanup:
+#         os.remove(file_path)
 
-    return combined_audio_filename
+#     return combined_audio_filename
 
 # Define the expandable sections
 musicgen_micro_blurb = """
@@ -269,10 +275,11 @@ with gr.Blocks() as iface:
     hidden_trigger = gr.Textbox(value="generate", visible=False)
 
     # Fixed click handlers - use hidden input for generate_drum_sample
-    generate_button.click(generate_drum_sample, inputs=[hidden_trigger], outputs=[drum_audio])
-    continue_drum_sample_button.click(continue_drum_sample, inputs=[drum_audio], outputs=[drum_audio])
-    generate_music_button.click(generate_music, inputs=[drum_audio, prompt_duration, musicgen_model, output_duration], outputs=[output_audio])
-    continue_button.click(continue_music, inputs=[output_audio, prompt_duration, musicgen_model, output_duration], outputs=continue_output_audio)
+    # Normal click connection
+    generate_button.click(generate_drum_sample, outputs=[drum_audio])
+    # continue_drum_sample_button.click(continue_drum_sample, inputs=[drum_audio], outputs=[drum_audio])
+    # generate_music_button.click(generate_music, inputs=[drum_audio, prompt_duration, musicgen_model, output_duration], outputs=[output_audio])
+    # continue_button.click(continue_music, inputs=[output_audio, prompt_duration, musicgen_model, output_duration], outputs=continue_output_audio)
 
 if __name__ == "__main__":
     iface.launch()