arioso.platforms.riffusion.adapter

Riffusion adapter using the riffusion library or diffusers fallback.

class arioso.platforms.riffusion.adapter.Adapter(config: dict)[source]

Riffusion adapter with dual backend support.

Tries the riffusion library first for full spectrogram-to-audio support. Falls back to the diffusers StableDiffusion pipeline which generates a spectrogram image and converts it to audio via inverse STFT.

generate(prompt: str, *, negative_prompt: str = '', seed: int | None = None, guidance: float = 7.0, num_steps: int = 50, audio_input_strength: float = 0.75, **kwargs) → Song[source]

Generate audio from a text prompt via spectrogram synthesis.

Parameters:

  • prompt – Text description of desired music.

  • negative_prompt – Elements to avoid in generation.

  • seed – Random seed for reproducibility.

  • guidance – Classifier-free guidance scale.

  • num_steps – Number of diffusion inference steps.

  • audio_input_strength – Denoising strength (0-1).

Returns:

A Song with audio_array populated.