ElevenLabs’ Scribe v3 Live labels each speaker during the call

The live model now labels who is speaking while the call runs, a first for ElevenLabs’ real-time transcription. It also changes turn-taking for voice agents. Current systems often wait for a set length of silence, which can cut off a caller who pauses halfway through reading out a number. Scribe v3 Live weighs the context of the conversation along with the audio to decide when someone has finished speaking, The Decoder reports. A multilingual mode handles callers who move between languages within one sentence, something ElevenLabs says its customers in India asked for with Hindi and English.

Developers can call both models through ElevenLabs’ API, where switching from Scribe v2 takes a single change to the model name. The live model also runs inside ElevenAgents, the company’s voice agent product. The accuracy figures are ElevenLabs’ own.

On our price sheet, checked 12 November, ElevenLabs’ Starter plan costs $6 a month for 30,000 credits with a commercial licence; the free plan is for personal use with attribution and bars paid ads.

Because speaker labels now arrive during the call, a live ElevenLabs transcript can show who said what as a conversation happens, so a sales or support agent can tell apart two people on the caller’s end while they talk.