tts.py

from __future__ import annotations
import base64
import os
import pathlib
import sys
from typing import Optional
from mistralai.client import Mistral

def synthesize_and_save_audio(
    input_text: str = "Hello!",
    voice_id: str = "",
    model: str = "voxtral-mini-tts-2603",
    api_key: str = "MISTRAL_API_KEY",
    output_path: str = "/tmp/voxtral.wav",
    response_format: str = "wav",
) -> int:

    client = Mistral(api_key=api_key)

    moderation_response = client.classifiers.moderate(
        model="mistral-moderation-2603",
        inputs=[input_text]
    )

    if moderation_response.results[0].categories["sexual"] or \
        moderation_response.results[0].categories["hate_and_discrimination"] or \
        moderation_response.results[0].categories["violence_and_threats"] or \
        moderation_response.results[0].categories["selfharm"] or \
        moderation_response.results[0].categories["jailbreaking"]:
        print("Input text blocked by moderation layer.", file=sys.stderr)
        return 3

    if "." in voice_id:
        print("Cloning voice from reference audio...")
        reference_path = pathlib.Path(voice_id).expanduser().resolve()
        if not reference_path.is_file():
            print(f"Reference audio not found: {reference_path}", file=sys.stderr)
            return 2

        with open(reference_path, "rb") as f:
            reference_bytes = f.read()
        reference_b64 = base64.b64encode(reference_bytes).decode("ascii")
        response = client.audio.speech.complete(
            model=model,
            input=input_text,
            ref_audio=reference_b64,
            response_format=response_format,
        )
    else:
        response = client.audio.speech.complete(
            model=model,
            input=input_text,
            voice_id=voice_id,
            response_format=response_format,
        )

    audio_bytes = base64.b64decode(response.audio_data)
    output_path_obj = pathlib.Path(output_path).expanduser()
    output_path_obj.write_bytes(audio_bytes)
    print(f"Wrote {len(audio_bytes)} bytes to {output_path_obj}")

    return 0