codemo/fish-speech-1

1from typing import Annotated, Literal, Optional2 3from pydantic import BaseModel, Field, conint4 5 6class ServeReferenceAudio(BaseModel):7    audio: bytes8    text: str9 10 11class ServeTTSRequest(BaseModel):12    text: str13    chunk_length: Annotated[int, conint(ge=100, le=300, strict=True)] = 20014    # Audio format15    format: Literal["wav", "pcm", "mp3"] = "wav"16    mp3_bitrate: Literal[64, 128, 192] = 12817    # References audios for in-context learning18    references: list[ServeReferenceAudio] = []19    # Reference id20    # For example, if you want use https://fish.audio/m/7f92f8afb8ec43bf81429cc1c9199cb1/21    # Just pass 7f92f8afb8ec43bf81429cc1c9199cb122    reference_id: str | None = None23    # Normalize text for en & zh, this increase stability for numbers24    normalize: bool = True25    mp3_bitrate: Optional[int] = 6426    opus_bitrate: Optional[int] = -100027    # Balance mode will reduce latency to 300ms, but may decrease stability28    latency: Literal["normal", "balanced"] = "normal"29    # not usually used below30    streaming: bool = False31    emotion: Optional[str] = None32    max_new_tokens: int = 102433    top_p: Annotated[float, Field(ge=0.1, le=1.0, strict=True)] = 0.734    repetition_penalty: Annotated[float, Field(ge=0.9, le=2.0, strict=True)] = 1.235    temperature: Annotated[float, Field(ge=0.1, le=1.0, strict=True)] = 0.736