arioso.platforms.stable_audio.adapter#

Stable Audio Open adapter using HuggingFace Diffusers.

Classes

Adapter(config)

Stable Audio Open adapter with lazy model loading.
class arioso.platforms.stable_audio.adapter.Adapter(config)[source]#

Bases: object

Stable Audio Open adapter with lazy model loading.

Uses diffusers.StableAudioPipeline for local inference.

generate(prompt, *, negative_prompt=None, duration=10.0, num_steps=200, guidance=7.0, seed=None, batch_size=1, sampler='dpmpp-3m-sde', audio_input=None, audio_input_strength=None, **kwargs)[source]#

Generate audio from a text prompt, optionally conditioned on input audio.

Parameters:

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

  • negative_prompt (str) – Text description of undesired characteristics.

  • duration (float) – Length in seconds.

  • num_steps (int) – Number of diffusion inference steps.

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

  • seed (int) – Random seed for reproducibility.

  • batch_size (int) – Number of waveforms to generate.

  • sampler (str) – Sampler type for the diffusion process.

  • audio_input – Optional input audio to condition on (audio-to-audio): a Song/AudioResult/bytes/path/(array, sample_rate)/NumPy waveform. Passed to the pipeline as initial_audio_waveforms – i.e. the model continues/initializes from this audio.

  • audio_input_strength (float) – Accepted for API symmetry but ignored – the diffusers StableAudioPipeline exposes no denoise/strength control (that is a stable-audio-tools/ComfyUI feature). A warning is emitted if a value is passed alongside audio_input.

Return type:

Song

Returns:

A Song with audio_array populated.