Add online serving to Stable Audio Diffusion and introduce v1/audio/generate endpoint (#1255)

This commit is contained in:
Ekagra Ranjan authored and GitHub committed 2026-03-09 13:36:46 -04:00
1 parent 761eff930a
commit 7a560492a2
16 files changed
+1860 -63

No files matched your search

@@ -85,6 +85,53 @@ class OpenAICreateSpeechRequest(BaseModel):
return self
class OpenAICreateAudioGenerateRequest(BaseModel):
"""Request model for audio generation via diffusion models (e.g. Stable Audio)."""
input: str = Field(
description="Text prompt describing the audio to generate",
)
model: str | None = None
response_format: Literal["wav", "pcm", "flac", "mp3", "aac", "opus"] = "wav"
speed: float | None = Field(
default=1.0,
ge=0.25,
le=4.0,
)
stream_format: Literal["sse", "audio"] | None = "audio"
audio_length: float | None = Field(
default=None,
description="Audio length in seconds",
)
audio_start: float | None = Field(
default=0.0,
description="Audio start time in seconds",
)
negative_prompt: str | None = Field(
default=None,
description="Negative prompt for classifier-free guidance",
)
guidance_scale: float | None = Field(
default=None,
description="Guidance scale for diffusion models",
)
num_inference_steps: int | None = Field(
default=None,
description="Number of inference steps",
)
seed: int | None = Field(
default=None,
description="Random seed for reproducibility",
)
@field_validator("stream_format")
@classmethod
def validate_stream_format(cls, v: str) -> str:
if v == "sse":
raise ValueError("'sse' is not a supported stream_format yet. Please use 'audio'.")
return v
class CreateAudio(BaseModel):
audio_tensor: np.ndarray
sample_rate: int = 24000