| """Hugging Face Inference Endpoints custom handler for Breeze TTS 2. |
| |
| POST JSON: |
| {"inputs": "Hello from Breeze.", "parameters": {"instruction": "A warm narrator."}} |
| {"input": "...", "instructions": "...", "voice": "..."} # OpenAI-ish aliases |
| |
| Returns WAV bytes as base64 plus sampling_rate so the toolkit and suite |
| wake path can both consume it. |
| """ |
| from __future__ import annotations |
|
|
| from typing import Any |
|
|
| from generate import sample_rate, synthesize |
|
|
|
|
| class EndpointHandler: |
| def __init__(self, path: str = "") -> None: |
| if path: |
| import os |
|
|
| os.environ.setdefault("BREEZE_MODEL_PATH", path) |
| |
| sample_rate() |
|
|
| def __call__(self, data: dict[str, Any]) -> dict[str, Any]: |
| payload = dict(data or {}) |
| parameters = payload.pop("parameters", None) or {} |
| if not isinstance(parameters, dict): |
| parameters = {} |
| text = ( |
| payload.pop("inputs", None) |
| or payload.pop("input", None) |
| or payload.pop("text", None) |
| or parameters.get("text") |
| or "" |
| ) |
| instruction = ( |
| payload.pop("instructions", None) |
| or payload.pop("instruction", None) |
| or payload.pop("voice", None) |
| or parameters.get("instruction") |
| or parameters.get("instructions") |
| or parameters.get("voice") |
| ) |
| cfg = payload.pop("cfg_scale", parameters.get("cfg_scale")) |
| seed = payload.pop("seed", parameters.get("seed", 42)) |
| wav = synthesize( |
| str(text), |
| instruction=str(instruction) if instruction else None, |
| cfg_scale=float(cfg) if cfg is not None else None, |
| seed=int(seed or 42), |
| ) |
| import base64 |
|
|
| return { |
| "audio": base64.b64encode(wav).decode("ascii"), |
| "sampling_rate": sample_rate(), |
| "content_type": "audio/wav", |
| } |
|
|
