Mistral releases Voxtral 3 — STT, TTS and a voice agent SDK in one

Rapid Orbit · 12 likes

voice agent SDK looks a lot like livekit agents with a mistral coat of paint

Isha Menon · 0 likes

not a bad thing, livekit got the abstractions right

Quiet Fern · 11 likes

the TTS is the surprise, its way more natural than i expected from mistral

Noah Meier · 7 likes

still a little flat on emotion imo but for agents thats fine

Cedar Runner · 10 likes

anyone know license for the bigger weights? last time it was research only

Tanvir Rahman · 4 likes

apache for small, the large one is api only as far as I can tell

Rohan Mehta · 8 likes

European option for voice is a big deal for companies with GDPR paranoia

Aoi Ito · 1 likes

paranoia is a strong word, its just legal reality here

Youssef Mahmoud · 1 likes

our legal team literally asked for this last week

Quiet Fern · 2 likes

price per minute vs elevenlabs? thats what my boss will ask first

Rapid River · 8 likes

roughly a third for the TTS based on the pricing page, but quality isnt identical

Luca Bianchi · 1 likes

tried the small one on my 4090, transcription latency under 200ms for short clips. pretty great

Grace Parker · 14 likes

what about accents? whisper always mangled my scottish mates

Urban Rider · 2 likes

scottish accent is the final boss of ASR, no model has won yet

Achieng Njeri · 0 likes

interruption handling is what makes or breaks voice agents. does it do barge-in well?

Valeria López · 13 likes

did a quick test, it stops talking within ~300ms when you interrupt. not bad

Mango Panda · 0 likes

still nowhere near a human who just ignores you and keeps talking lol