examples/inference.py

#!/usr/bin/env python3
import argparse
from pathlib import Path

import torch
from scipy.io.wavfile import write
from transformers import AutoTokenizer, VitsModel


def parse_args() -> argparse.Namespace:
    parser = argparse.ArgumentParser(description="Generate Khmer speech with the local MMS TTS model.")
    parser.add_argument("--model", default=str(Path(__file__).resolve().parents[1]))
    parser.add_argument("--text", required=True)
    parser.add_argument("--output", default="khmer_tts.wav")
    return parser.parse_args()


def main() -> None:
    args = parse_args()
    tokenizer = AutoTokenizer.from_pretrained(args.model)
    model = VitsModel.from_pretrained(args.model)
    model.eval()

    inputs = tokenizer(args.text, return_tensors="pt")
    with torch.no_grad():
        waveform = model(**inputs).waveform.squeeze().cpu().numpy()

    write(args.output, rate=model.config.sampling_rate, data=waveform)
    print(f"Wrote {args.output} at {model.config.sampling_rate} Hz")


if __name__ == "__main__":
    main()