| import sys, json, traceback |
| from pathlib import Path |
| import numpy as np, soundfile as sf, torch |
| from qwen_tts import Qwen3TTSModel |
| B = Path("/marimo/ai_studio") |
| model = Qwen3TTSModel.from_pretrained(str(B/"models"/"lyra-voice-cantonese"), |
| device_map="cuda", dtype=torch.bfloat16) |
| langs = model.get_supported_languages(); spks = model.get_supported_speakers() |
| print("LANGS:", langs, flush=True) |
| print("SPEAKERS:", spks, flush=True) |
| default_spk = (spks[0] if isinstance(spks, (list, tuple)) and spks else spks) |
| spec = json.loads(Path(sys.argv[1]).read_text()) |
| for it in spec: |
| try: |
| spk = it.get("speaker") or default_spk |
| kw = dict(text=it["text"], speaker=spk) |
| if it.get("language"): kw["language"] = it["language"] |
| wavs, sr = model.generate_custom_voice(**kw) |
| out = B/"assets/audio"/it["name"] |
| sf.write(str(out), np.asarray(wavs[0]), sr) |
| print("WROTE", out.name, f"{len(np.asarray(wavs[0]))/sr:.2f}s", flush=True) |
| except Exception as e: |
| print("FAIL", it.get("name"), type(e).__name__, e, flush=True) |
| traceback.print_exc() |
| print("TTS_DONE") |
|
|
