tts.py

"""Pydantic models for the TTS (text-to-speech) proxy endpoint.

Defines the request shape shared by mobile and desktop clients.
"""

from typing import Optional
from pydantic import BaseModel, ConfigDict, Field

DEFAULT_VOICE_ID = "BAMYoBHLZM7lJgJAmFz0"  # Sloane
DEFAULT_MODEL_ID = "eleven_turbo_v2_5"
DEFAULT_OUTPUT_FORMAT = "mp3_44100_128"


class TtsVoiceSettings(BaseModel):
    stability: Optional[float] = None
    similarity_boost: Optional[float] = None
    style: Optional[float] = None
    use_speaker_boost: Optional[bool] = None


class TtsSynthesizeRequest(BaseModel):
    # `model_id` collides with pydantic's `model_` protected namespace; disable.
    model_config = ConfigDict(protected_namespaces=())

    # Upper bound mirrors _TTS_REQUEST_CHAR_LIMIT in routers/tts.py (and the Rust backend), so an
    # oversized request is rejected at the schema/contract boundary rather than only at runtime.
    text: str = Field(..., min_length=1, max_length=5000)
    voice_id: str = DEFAULT_VOICE_ID
    model_id: str = DEFAULT_MODEL_ID
    output_format: str = DEFAULT_OUTPUT_FORMAT
    voice_settings: Optional[TtsVoiceSettings] = None