# -*- coding: utf-8 -*-
"""Ryan 音色中文发音冒烟测试：决定 simon/rhythm 男声旁白选型。"""
import os
os.environ.setdefault("HF_HOME", "/home/ubuntu/SoundGame/models/hf")
os.environ.setdefault("HF_ENDPOINT", "https://hf-mirror.com")
import numpy as np
import soundfile as sf
import torch
from qwen_tts import Qwen3TTSModel

model = Qwen3TTSModel.from_pretrained(
    "Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice",
    device_map="cuda:0",
    dtype=torch.bfloat16,
)
texts = [
    "欢迎来到音序回响。我会用六种声音演奏一段序列，请你记住顺序。",
    "听好了：上行音，向上滑；下行音，向下滑。预备，开始！",
    "零一二三四五六七八九十",
]
wavs, sr = model.generate_custom_voice(
    text=texts,
    language=["Chinese"] * len(texts),
    speaker=["Ryan"] * len(texts),
    instruct=["热情活力的中文游戏解说语气，吐字清晰"] * len(texts),
)
os.makedirs("/home/ubuntu/SoundGame/tools/samples", exist_ok=True)
for i, wav in enumerate(wavs):
    wav = np.asarray(wav, dtype=np.float32)
    sf.write(f"/home/ubuntu/SoundGame/tools/samples/ryan_zh_{i}.wav", wav, sr)
    print(f"ryan_zh_{i}: {len(wav)/sr:.2f}s peak={np.max(np.abs(wav)):.3f}")
print("done")