Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
peturdarri
searching Neon…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
7 ms
·
1.
▲
by
peturdarri
1y ago
You're right, this is not solvable with regular LLMs. It's not possible to mimic natural conversational rhythm with a separate LLM generating text, a separate text-to-speech generating audio, and a separate VAD determining when to
2.
▲
by
peturdarri
3y ago
According to the technical paper ( https://goo.gle/GeminiPaper ), Gemini Nano-1, the smallest model at 1.8B parameters, beats Whisper large-v3 and Google's USM at automatic speech recognition. That's very impressive