Scribe v3 and v3 Live (elevenlabs.io)

ElevenLabs has unveiled Scribe v3 and Scribe v3 Live, two enhanced Speech to Text models tailored for challenging audio environments like meetings and customer interactions. Scribe v3 significantly reduces word errors by 27% compared to its predecessor, Scribe v2, across 102 languages and features the best speaker labeling performance. Meanwhile, Scribe v3 Live offers real-time transcription with a remarkable 43% reduction in errors, incorporating speaker diarization for the first time, which accurately identifies speakers in dynamic conversations. These advancements are critical for the AI/ML community as they address the perennial challenges of real-world audio, enhancing the accuracy and usability of transcription across various applications, from clinical documentation to customer service. Key improvements include better handling of overlapping dialogue, multilingual capabilities that automatically adapt to language switches, and more precise formatting controls, allowing users to dictate how outputs—like numbers and jargon—are presented. As organizations increasingly rely on voice technologies, the effectiveness of these models could transform how businesses utilize AI-driven transcription solutions, ensuring conversations are documented correctly from the start, ultimately saving time and resources.