from Hacker News

Elevenlabs Conversational AI 2.0

by killion on 6/1/25, 6:48 PM with 11 comments

• by miles on 6/1/25, 7:06 PM

  Given their excellent record, it's strange that ElevenLabs would make a basic transcription mistake in the embedded introduction video[1].

  At around 30 seconds in[2], an AI agent types and says, "Hi! Welcome to our design studio." to which a customer responds「すみません、日本語で話してもいいですか？」(Sorry, may I speak Japanese?).

  The AI agent then says 「もちろんです。日本語に切り替えましょう。」(Of course. Let's switch to Japanese.) but types「ご心配なく。数量は柔軟に対応できます。」which means "No worries. We can be flexible with the quantity."

  [1] https://www.youtube.com/watch?v=TlclS4wLWgY

  [2] https://imgur.com/18U9uUU

• by gorgoiler on 6/1/25, 10:06 PM

  The play button widget will read the article aloud. Alas, it called the product “conversational AI two dot zero comma” and the company “vertical bar elevenlabs”.

  Their TTS is good but unfortunately the exemplar widget is not nearly up to the same quality.

• by giaccoangelo on 6/1/25, 8:00 PM

  a lot of awesome features in this ship, comment your feedback for me:)

• by wild_egg on 6/1/25, 11:44 PM

  What does SOTA tool use with voice agents look like these days? Any providers have MCP support?

  It'd be incredible to get to a point where we can have natural conversations and the AI is running tools in the background and keeping tabs on things.

• by ldenoue on 6/1/25, 8:44 PM

  How is the turn detection working? LLM prompting or a special AI audio plus text model?