index out of range error fix

app.py

@@ -31,7 +31,13 @@ def create_slices(song, sr, slice_duration, bpm, num_slices=5):
     slices.append(first_slice_waveform)
     
     for i in range(1, num_slices):
-        random_start = random.choice(range(int(slice_duration * sr), int(song_length * sr), int(4 * 60 / bpm * sr)))
+        possible_start_indices = list(range(int(slice_duration * sr), int(song_length * sr), int(4 * 60 / bpm * sr)))
+        if not possible_start_indices:
+            # If there are no valid start indices, duplicate the first slice
+            slices.append(first_slice_waveform)
+            continue
+        
+        random_start = random.choice(possible_start_indices)
         slice_end = random_start + int(slice_duration * sr)
         
         if slice_end > song_length * sr:
@@ -65,105 +71,110 @@ def calculate_duration(bpm, min_duration=29, max_duration=30):
     return duration
 
 def generate_music(seed, use_chords, chord_progression, prompt_duration, musicgen_model, num_iterations, bpm):
-    if seed == "":
-        seed = random.randint(1, 10000)
-
-    ml = MusicLangPredictor('musiclang/musiclang-v2')
-
-    try:
-        seed = int(seed)
-    except ValueError:
-        seed = random.randint(1, 10000)
-
-    nb_tokens = 2048
-    temperature = 0.9
-    top_p = 1.0
-
-    if use_chords and chord_progression.strip():
-        score = ml.predict_chords(
-            chord_progression,
-            time_signature=(4, 4),
-            temperature=temperature,
-            topp=top_p,
-            rng_seed=seed
-        )
-    else:
-        score = ml.predict(
-            nb_tokens=nb_tokens,
-            temperature=temperature,
-            topp=top_p,
-            rng_seed=seed
-        )
-
-    midi_filename = f"output_{seed}.mid"
-    wav_filename = midi_filename.replace(".mid", ".wav")
-
-    score.to_midi(midi_filename, tempo=bpm, time_signature=(4, 4))
-
-    subprocess.run(["fluidsynth", "-ni", "font.sf2", midi_filename, "-F", wav_filename, "-r", "44100"])
-
-    # Load the generated audio
-    song, sr = torchaudio.load(wav_filename)
-    song = song.to(device)
-
-    # Use the user-provided BPM value for duration calculation
-    duration = calculate_duration(bpm)
-
-    # Create slices from the song using the user-provided BPM value
-    slices = create_slices(song, sr, 35, bpm, num_slices=5)
-
-    # Load the model
-    model_name = musicgen_model.split(" ")[0]
-    model_continue = MusicGen.get_pretrained(model_name)
-
-    # Setting generation parameters
-    model_continue.set_generation_params(
-        use_sampling=True,
-        top_k=250,
-        top_p=0.0,
-        temperature=1.0,
-        duration=duration,
-        cfg_coef=3
-    )
-
-    all_audio_files = []
-
-    for i in range(num_iterations):
-        slice_idx = i % len(slices)
-        
-        print(f"Running iteration {i + 1} using slice {slice_idx}...")
-        
-        prompt_waveform = slices[slice_idx][..., :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()  # Move the output tensor back to CPU
-        
-        # Make sure the output tensor has at most 2 dimensions
-        if len(output.size()) > 2:
-            output = output.squeeze()
-        
-        filename_without_extension = f'continue_{i}'
-        filename_with_extension = f'{filename_without_extension}.wav'
-        
-        audio_write(filename_with_extension, output, model_continue.sample_rate, strategy="loudness", loudness_compressor=True)
-        all_audio_files.append(f'{filename_without_extension}.wav.wav')  # Assuming the library appends an extra .wav
-
-    # Combine all audio files
-    combined_audio = AudioSegment.empty()
-    for filename in all_audio_files:
-        combined_audio += AudioSegment.from_wav(filename)
-
-    combined_audio_filename = f"combined_audio_{seed}.mp3"
-    combined_audio.export(combined_audio_filename, format="mp3")
-
-    # Clean up temporary files
-    os.remove(midi_filename)
-    os.remove(wav_filename)
-    for filename in all_audio_files:
-        os.remove(filename)
-
-    return combined_audio_filename
+    while True:
+        try:
+            if seed == "":
+                seed = random.randint(1, 10000)
+
+            ml = MusicLangPredictor('musiclang/musiclang-v2')
+
+            try:
+                seed = int(seed)
+            except ValueError:
+                seed = random.randint(1, 10000)
+
+            nb_tokens = 1024
+            temperature = 0.9
+            top_p = 1.0
+
+            if use_chords and chord_progression.strip():
+                score = ml.predict_chords(
+                    chord_progression,
+                    time_signature=(4, 4),
+                    temperature=temperature,
+                    topp=top_p,
+                    rng_seed=seed
+                )
+            else:
+                score = ml.predict(
+                    nb_tokens=nb_tokens,
+                    temperature=temperature,
+                    topp=top_p,
+                    rng_seed=seed
+                )
+
+            midi_filename = f"output_{seed}.mid"
+            wav_filename = midi_filename.replace(".mid", ".wav")
+
+            score.to_midi(midi_filename, tempo=bpm, time_signature=(4, 4))
+
+            subprocess.run(["fluidsynth", "-ni", "font.sf2", midi_filename, "-F", wav_filename, "-r", "44100"])
+
+            # Load the generated audio
+            song, sr = torchaudio.load(wav_filename)
+            song = song.to(device)
+
+            # Use the user-provided BPM value for duration calculation
+            duration = calculate_duration(bpm)
+
+            # Create slices from the song using the user-provided BPM value
+            slices = create_slices(song, sr, 35, bpm, num_slices=5)
+
+            # Load the model
+            model_name = musicgen_model.split(" ")[0]
+            model_continue = MusicGen.get_pretrained(model_name)
+
+            # Setting generation parameters
+            model_continue.set_generation_params(
+                use_sampling=True,
+                top_k=250,
+                top_p=0.0,
+                temperature=1.0,
+                duration=duration,
+                cfg_coef=3
+            )
+
+            all_audio_files = []
+
+            for i in range(num_iterations):
+                slice_idx = i % len(slices)
+                
+                print(f"Running iteration {i + 1} using slice {slice_idx}...")
+                
+                prompt_waveform = slices[slice_idx][..., :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()  # Move the output tensor back to CPU
+                
+                # Make sure the output tensor has at most 2 dimensions
+                if len(output.size()) > 2:
+                    output = output.squeeze()
+                
+                filename_without_extension = f'continue_{i}'
+                filename_with_extension = f'{filename_without_extension}.wav'
+                
+                audio_write(filename_with_extension, output, model_continue.sample_rate, strategy="loudness", loudness_compressor=True)
+                all_audio_files.append(f'{filename_without_extension}.wav.wav')  # Assuming the library appends an extra .wav
+
+            # Combine all audio files
+            combined_audio = AudioSegment.empty()
+            for filename in all_audio_files:
+                combined_audio += AudioSegment.from_wav(filename)
+
+            combined_audio_filename = f"combined_audio_{seed}.mp3"
+            combined_audio.export(combined_audio_filename, format="mp3")
+
+            # Clean up temporary files
+            os.remove(midi_filename)
+            os.remove(wav_filename)
+            for filename in all_audio_files:
+                os.remove(filename)
+
+            return combined_audio_filename
+        except IndexError:
+            # Retry with a new random seed if an IndexError is raised
+            seed = random.randint(1, 10000)
 
 # Check if CUDA is available
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")