Last week NEURA got a face. This week it got ears and a voice. That was always the gap that bugged me: a companion you can see but not actually talk to is just a screensaver with opinions.
The first problem was hearing. I had been running a small speech model locally, everything on my own hardware, which sounded clean until it kept tripping over simple words. Nothing kills the feeling of talking to something faster than watching it mangle "what time is it." So I moved transcription to ElevenLabs Scribe. It catches full sentences in about half a second, and it catches them right.
Then the voice. I added an ElevenLabs voice so NEURA answers out loud instead of printing text, and it lands in a natural tone instead of the flat robot cadence I was bracing for. Claude still does the thinking behind every reply, so what you hear is a real answer, just spoken.
To go hands free I paired a Bluetooth headset with both a microphone and a speaker, so I can talk from across the room with nothing to reach for. I also made it keep listening for a few seconds after it stops talking, so you can ask a follow-up without repeating its name. Small detail, big difference in how it feels.
Then the moment. I asked it a real question and heard it answer out loud for the first time. I have built a lot of quiet software over the years, and hearing this one talk back landed differently. Next week I want to tighten the round trip and let it handle interruptions, but for now it hears me and it talks. That is enough of a win to close the week on.