app/api/speech.py

from fastapi import APIRouter, Depends, Response

from app.auth import require_bearer_auth
from app.config import Settings, get_settings
from app.errors import OpenAICompatibleError
from app.schemas import SpeechRequest
from app.services.audio_formatter import format_audio
from app.services.engine_registry import EngineRegistry, get_engine_registry
from app.utils.text import ensure_max_length

router = APIRouter(prefix="/v1")


@router.post("/audio/speech", dependencies=[Depends(require_bearer_auth)])
def create_speech(
    payload: SpeechRequest,
    settings: Settings = Depends(get_settings),
    registry: EngineRegistry = Depends(get_engine_registry),
) -> Response:
    ensure_max_length(payload.input, settings.max_input_length)

    engine = registry.get_engine(payload.model)

    if not engine.supports_voice(payload.voice):
        raise OpenAICompatibleError(
            status_code=400,
            message=f"Unsupported voice '{payload.voice}'.",
            param="voice",
            code="unsupported_voice",
        )

    response_format = payload.response_format or settings.default_response_format
    speed = payload.speed if payload.speed is not None else settings.default_speed
    quality = payload.quality or settings.default_quality

    synthesis = engine.synthesize(
        text=payload.input,
        voice=payload.voice,
        speed=speed,
        quality=quality,
        model_name=payload.model,
        lang=settings.default_language,
    )
    audio_bytes, media_type = format_audio(
        waveform=synthesis.waveform,
        sample_rate=synthesis.sample_rate,
        response_format=response_format,
    )
    return Response(content=audio_bytes, media_type=media_type)