| 123456789101112131415161718192021222324252627282930313233343536 |
- from typing import Annotated, Literal, Optional
- from pydantic import BaseModel, Field, conint
- class ServeReferenceAudio(BaseModel):
- audio: bytes
- text: str
- class ServeTTSRequest(BaseModel):
- text: str
- chunk_length: Annotated[int, conint(ge=100, le=300, strict=True)] = 200
- # Audio format
- format: Literal["wav", "pcm", "mp3"] = "wav"
- mp3_bitrate: Literal[64, 128, 192] = 128
- # References audios for in-context learning
- references: list[ServeReferenceAudio] = []
- # Reference id
- # For example, if you want use https://fish.audio/m/7f92f8afb8ec43bf81429cc1c9199cb1/
- # Just pass 7f92f8afb8ec43bf81429cc1c9199cb1
- reference_id: str | None = None
- seed: int | None = None
- use_memory_cache: Literal["on-demand", "never"] = "never"
- # Normalize text for en & zh, this increase stability for numbers
- normalize: bool = True
- mp3_bitrate: Optional[int] = 64
- opus_bitrate: Optional[int] = -1000
- # Balance mode will reduce latency to 300ms, but may decrease stability
- latency: Literal["normal", "balanced"] = "normal"
- # not usually used below
- streaming: bool = False
- max_new_tokens: int = 1024
- top_p: Annotated[float, Field(ge=0.1, le=1.0, strict=True)] = 0.7
- repetition_penalty: Annotated[float, Field(ge=0.9, le=2.0, strict=True)] = 1.2
- temperature: Annotated[float, Field(ge=0.1, le=1.0, strict=True)] = 0.7
|
