Voice AIs handle speaker turns & interruptions with Voice Activity Detection. VAD is brittle and will trigger due to background noise, creating frequent hiccups. Moshi gets rid of it completely, so you can use it in the most chaotic settings. I myself couldn't hear Moshi here 😅

@Voice I loved talking to moshi while it talked to me too at the same time. By the end of our combo we were only half talking and just exchanging appropriate sounding grunts :)

The perks of nice Parisian headquarters!😅

This is a great demo and I sensed something different with data handling recently when I was using mobile outside my radius. I had one bar. Moshi was able to adjust the buffer speed and I think was "piecing me" together differently. IT was incredibly effective. Thank you./1

Great work! But any idea why Moshi seems to have hiccups and delays like in the following chat? Is it because of the accent? This was between and my own s2s bot, and this was actually one of the better examples in terms of hiccups.

What hardware is the inference running in?