audiocraft-audio-generation

Load a pretrained model with MusicGen.get_pretrained('facebook/musicgen-small'), set generation parameters like duration and temperature, then call model.generate with a list of text descriptions. Save the output tensor with torchaudio.save at 32000 Hz.

MusicGen generates music from text descriptions and supports melody conditioning, while AudioGen generates sound effects and environmental audio. MusicGen outputs at 32kHz and AudioGen at 16kHz.

MusicGen-small needs about 2GB VRAM in FP16, medium needs about 4GB, and large needs about 8GB. If you hit CUDA out-of-memory errors, use a smaller model, reduce duration, or clear the CUDA cache between generations.

Yes, use the facebook/musicgen-melody model and call generate_with_chroma with your text descriptions, a melody waveform, and its sample rate. The melody should be mono and resampled to 32kHz.

Silent output usually means the audio needs normalization to a target loudness. Distortion or clipping can be fixed by scaling amplitudes below 1.0, applying soft clipping, or lowering the generation temperature for more controlled output.

Use Stable Audio for longer commercial music generation and Bark for text-to-speech combined with music or sound effects. AudioCraft is best for controllable text-to-music, melody conditioning, and sound effect generation.