ElevenLabs launches speech models for noisy and real-time conversations
ElevenLabs has released Scribe v3 and Scribe v3 Live, two speech-to-text models designed for difficult audio, including overlapping speakers and background noise. Scribe v3 is intended for recorded files, while Scribe v3 Live is aimed at low-latency applications. Both are available through the ElevenAPI developer interface and the ElevenAgents platform.
The company says Scribe v3 reduces word errors by about 25 percent compared with its predecessor across 102 languages. It also claims improved German recognition and more reliable speaker diarization in larger conversations. Artificial Analysis benchmarks placed Scribe v3 among the most accurate tested models, narrowly ahead of offerings from Microsoft, Mistral, xAI, Google and OpenAI, while an Alibaba preview model ranked higher. The test recorded a price of $3.33 per 1,000 minutes, and showed that the model was slower and more expensive than some rivals.
Scribe v3 Live uses conversational context as well as audio to distinguish a genuine end of speech from a short pause, reducing interruptions by voice agents. Developers can add specialized terms, define number and date formatting, and enable multilingual transcription. Enterprise options include EU-isolated servers and zero-retention processing. ElevenLabs also announced expanded work with Nvidia, an alliance with McKinsey and a marketplace for voice agents.
Written by Sakhanda Wire AI from reporting by The Decoder. Facts, quotes and full context are in the original article below.