Audio & Speech

Today's Stories

1. 1

   MeetTwins: AI bot attends Google Meet calls for you

   Aditya Shinde is recruiting beta users for MeetTwins, an AI assistant built with Sarvam's AI models and a Simli avatar that attends Google Meet calls and answers only pre-approved questions. The tool lets workers skip routine check-ins, but colleagues who dislike it can type "/stop" to eject the bot immediately, and the operator gets an email summary afterward.

   Whether labeling the bot as AI actually "builds trust" is Shinde's claim, not a proven result; the product is still in development with no launch date given.

2. 2

   ElevenLabs employees cash out at $22 billion valuation

   ElevenLabs ran a $300 million tender offer letting employees sell vested shares to investors, co-led by Wellington and T. Rowe Price, at a $22 billion valuation. ElevenLabs is using employee liquidity as a retention tool, part of a growing trend among fast-growing AI startups aiming to keep staff from leaving for competitors.

   The transaction's backers are institutional investors who plan to hold the stock after an eventual IPO, so the test is whether that backing holds through a public listing.

3. 3

   Hugging Face Open TTS Leaderboard cuts eval from weeks to hours

   Hugging Face launched the Open TTS Leaderboard, ranking models by word error rate (WER/CER), H200 GPU speed measures, and voice-cloning speaker similarity, cutting evaluation from a couple weeks to a couple hours. This gives open-source TTS models a standard, scalable way to be compared, which may help address their underrepresentation on human-vote arena leaderboards.

   The leaderboard relies on objective metrics like WER and speaker similarity, which may not fully capture naturalness or listener preference. Watch for community votes and feedback to shape future evaluations.

4. 4

   Claude Code AWS cost probe tops GENIEE Tech Blog reads

   GENIEE's tech blog team named its three most-read posts from September 2025 to September 2026, led by an SRE team that used Claude Code with AWS CLI and Cost Explorer to find tens of millions of yen in annual savings. The top post shows AI being used for investigation and analysis, not just coding, and the blog's three most-read pieces were all AI-related, suggesting that practical usage reports are what readers want.

   The scale of that saving figure hinges on how much of the identified potential is actually implemented, and the other two picks—a Whisper/pyannote transcription project and an engineer leader's AI tool stack—show where readers' attention is heading.

5. 5

   Local Speech-to-Speech: OpenAI Realtime API experience verified

   A developer verified how far a Speech-to-Speech experience like OpenAI Realtime API could be reproduced locally, using Hugging Face's speech-to-speech models. Running STT, LLM and TTS together on one GPU server, driven only by a Windows PC's mic and speakers, keeps the voice loop off cloud APIs.

   The stated goal is a system-level check of the local setup, not a benchmark of individual models, so the value hinges on how the full cascade performs.

6. 6

   Tokyo court dismisses Kenjiro Tsuda AI voice case

   The Tokyo District Court dismissed Kenjiro Tsuda's demand that TikTok remove AI voice-clone videos, citing prior deletion. Judge Aya Takahashi recognized voice as a protected publicity right. The ruling sets a precedent that voice can be protected under publicity rights even when the specific removal demand fails.

   Whether future cases proceed when videos remain online, and how the Justice Ministry's August guidelines on voice rights are applied.