TLDR: ElevenLabs has launched Scribe v2 Realtime, an advanced AI speech-to-text model capable of delivering human-quality live transcription in under 150 milliseconds. The model supports over 90 languages globally, including 11 Indian languages such as Kannada, Hindi, Tamil, and Telugu, achieving 93.5% accuracy on the FLEURS benchmark. It is designed for developers and enterprises building real-time voice applications, offering features like negative latency prediction and India-specific data residency.
ElevenLabs, a prominent innovator in voice AI, has officially launched Scribe v2 Realtime, marking a significant leap forward in speech-to-text technology. This new model is engineered to provide human-quality live transcription with remarkable speed, processing audio in under 150 milliseconds, setting a new industry benchmark for real-time communication.
Scribe v2 Realtime boasts extensive language support, covering more than 90 languages worldwide. Crucially for the Indian market, it includes support for 11 Indian languages, namely Hindi, Tamil, Malayalam, Kannada, Telugu, Gujarati, Odia, Bengali, Marathi, Punjabi, and Sindhi. The company reports an impressive 93.5% accuracy on the FLEURS benchmark across 30 European and Asian languages, underscoring its robust performance in diverse linguistic environments.
This advanced model is primarily aimed at developers and enterprises looking to build sophisticated voice assistants, efficient meeting transcription tools, and reliable live captioning applications. Its enterprise applications are broad, ranging from real-time transcription of customer calls for compliance monitoring and voice AI agents to live medical dictation, clinical documentation, and accessibility captions for educational and media platforms.
Key features of Scribe v2 Realtime include negative latency prediction for next-word and punctuation, text conditioning, and Voice Activity Detection (VAD) for automatic speech segmentation. It also offers manual commit controls for enhanced streaming performance, domain-specific custom vocabulary for improved accuracy in specialized contexts, smart speaker diarization, and precise timestamps. For sensitive workloads, a zero-retention mode is available, ensuring data privacy.
Also Read:
- ElevenLabs Unveils Enhanced Enterprise AI Agent Capabilities at Inaugural Summit
- TIME Magazine Introduces AI Agent to Revolutionize Reader Engagement
ElevenLabs has also prioritized global and regional compliance, offering data residency options in India to meet local data regulations. The model adheres to stringent enterprise security and privacy standards, including SOC 2, ISO 27001, PCI DSS L1, HIPAA, and GDPR. Scribe v2 Realtime is readily available today through the ElevenLabs API, allowing seamless integration into various applications, and can be directly deployed within ElevenLabs Agents, facilitating the creation of more natural and responsive conversational AI systems for support and sales workflows.
