Guardrails removed spam, off-topic, unclear, or duplicate replies.
Ask a question below.
Published answers will appear here.
Next experiment: found an old rotary phone in my basement and after some fiddling with the wires I am getting a tone from the Fritzbox. Now let's see if I can get a connection to a local Canary-Qwen model.
So here's the current state - both voice-to-text, end-of-speech detection and text-to-voice are local models but I let Claude Sonnet write the replies, which does take the longest. There is clearly room for improvement.
Oh, it already works. 1 second latency is adding some awkward pauses and Canary-Qwens' strengths seem to be more on the audio transcription side than on having an intelligent conversation, but then again, having it all local is a nice touch.
Guardrails removed spam, off-topic, unclear, or duplicate replies.
Ask a question below.
Published answers will appear here.