ElevenLabs has released Scribe v3 and Scribe v3 Live speech recognition models

ElevenLabs has introduced two new speech recognition models — Scribe v3 and Scribe v3 Live. They are designed for accurate audio-to-text transcription in environments with background noise and overlapping speech.

The new models are aimed at use in voice agents. The Scribe v3 Live version utilizes conversational context, which helps prevent situations where AI agents prematurely interrupt callers mid-sentence.

Independent benchmarks from Artificial Analysis confirm that the base Scribe v3 demonstrates high accuracy, slightly outperforming similar solutions from OpenAI, Google, and Microsoft.

However, the increased accuracy has impacted performance and cost. Competitive solutions from Microsoft and xAI run several times faster and are roughly half the price per volume of processing.

The arrival of these models provides developers of conversational interfaces with a tool to improve dialogue quality by leveraging context, despite the higher resource requirements.