ElevenLabs unveils Scribe V2

Scribe V2 is a model designed for large-scale batch transcription, subtitle, and caption generation. Compared to the previous v1, it handles longer audio, pauses, tone changes, and long silence sections more stably and accurately, and it recorded the lowest word error rate (WER) on FLEURS, an industry-standard benchmark.

Key features include the following:

• Keyterm Prompting: Uses context to accurately transcribe up to 100 words or phrases, effectively handling technical terms or brand names.
• Entity Detection: Automatically detects 56 categories including personal information (PII), medical data, and payment information, and provides precise timestamps.
• Automatic Multilingual Transcription: Even when a single file contains multiple languages, it automatically detects the language and transcribes without separate configuration.

For enterprise workflows, it supports Smart Diarization, precise word-level timestamps, and Dynamic Audio Tagging that detects non-speech events such as laughter or footsteps. It is also optimized for enterprise environments by complying with various security regulations such as SOC 2, ISO 27001, and HIPAA.

Scribe V2 is now available immediately through ElevenLabs Studio and the API.