arioso.platforms.musicgen.adapter#

MusicGen adapter using audiocraft (preferred) or transformers.

Supports text-to-music generation and, via facebook/musicgen-melody, melody-conditioned generation (the melody affordance / arioso.enhance stage): the output follows the pitch contour of an input audio while taking its style from the text prompt.

Classes

Adapter(config)

MusicGen adapter with lazy model loading.
class arioso.platforms.musicgen.adapter.Adapter(config)[source]#

Bases: object

MusicGen adapter with lazy model loading.

Tries audiocraft first; if unavailable, falls back to the HuggingFace transformers pipeline.

generate(prompt, *, duration=8.0, temperature=1.0, top_k=250, top_p=0.0, guidance=3.0, model='facebook/musicgen-small', melody=None, **kwargs)[source]#

Generate music from a text prompt, optionally following a melody.

Parameters:

  • prompt (str) – Text description of desired music.

  • duration (float) – Length in seconds.

  • temperature (float) – Sampling randomness.

  • top_k (int) – Top-k sampling parameter.

  • top_p (float) – Top-p nucleus sampling.

  • guidance (float) – Classifier-free guidance scale.

  • model (str) – Model variant name (e.g. ‘facebook/musicgen-small’). When a melody is supplied and model isn’t already a melody variant, it is switched to facebook/musicgen-melody.

  • melody – Optional input audio whose pitch contour the output follows (a Song/AudioResult/bytes/path/(array, sample_rate)/waveform). This is the arioso.enhance path for musicgen.

Return type:

Song

Returns:

A Song with audio_array populated.