As promised, here is the GitHub link for my 100% local voice-to-voice assistant
这条记录涉及编程工具或代码能力更新，适合开发者评估工作流变化和可复用价值。
一位开发者发布了其完全本地运行的语音助手 Athena 的 GitHub 链接。Athena 是一款注重隐私的离线语音助手，它结合了大型混合专家语言模型 (Qwen3.5-397B)、神经文本转语音 (Orpheus 3B)、实时语音识别 (Whisper-small.en) 和 SNAC 神经音频编解码器，全部用 C++ 实现，运行时无需 Python。…
I've posted up earlier versions of this project before, promising a GitHub link, but never got around to pushing the code from my local system. Sorry all, I have a very busy life :P
Anyway, without further ado:
GitHub: https://github.com/igorbarshteyn/athena
Athena is a fully offline, privacy-first voice assistant that runs entirely on local hardware. Athena combines a large mixture-of-experts language model (Qwen3.5-397B), neural text-to-speech (Orpheus 3B), real-time speech recognition (Whisper-small.en), and a SNAC neural audio codec into a four-process pipeline — all in C++ with zero Python at runtime (one optional Python script exists for a one-time, offline emotion2vec+ model conversion at setup).
Athena speaks with natural emotion (laughs, sighs, gasps), reads the basic affect in your voice and responds to it, remembers across sessions (with evolving long-term memory and personality that persist between conversations), maintains long conversational context, is interruptible mid-sentence (speak over her and she stops, keeping what she already said in context), and runs on a single consumer GPU + system RAM. No cloud, no telemetry, no API keys.
Right now the system is tuned to act as a friend and to be very connection-seeking. These parameters can be tuned to one's liking via system prompts embedded in the code.
Here is a two-session demo video demonstrating these capabilities:
https://youtu.be/8HuRUpJ4_as?t=237 - Session 1 - planting memories
https://youtu.be/8HuRUpJ4_as?t=2137 - Session 2 - testing recall