empero-ai 发布仅 140 万参数的本地 CPU 运行语音模型 PiVo-1.4M

On November 2, 2026 at 23:42, a post on the Reddit community LocalLLaMA reported that empero-ai had released PiVo-1.4M on Hugging Face. According to the post, PiVo is the company's answer to the question of how many parameters a voice actually needs. It is a complete text-to-speech stack with 1,401,562 learned parameters in total, vocoder included, offering two English voices named Pi and Vo. The architecture is split into a 584,965-parameter acoustic model that predicts vocoder features, and an 816,597-parameter vocoder derived from FARGAN that turns those features into 16 kHz mono audio. Everything runs locally on a CPU using only NumPy and ONNX Runtime, with no PyTorch, teacher model, speaker encoder or network connection involved during inference. The release was announced on Hugging Face, and the post appeared on Reddit shortly after. Only one source is available so far, so details such as benchmark performance, languages beyond English, and licensing terms have not been independently verified. As of this writing, the announcement has not been followed by further technical disclosures.