Archive — US — techcrunch
ElevenLabs is launching its own speech-to-text model
ElevenLabs, an AI startup valued at $3.3 billion and known for audio generation, launches its speech-to-text model 'Scribe' following a $180 million funding round. Scribe supports over 99 languages with impressive accuracy, aiming to compete with existing models from Google, OpenAI, and others. Currently, it only transcribes pre-recorded audio and is priced at $0.40 per hour.
Key points
- ElevenLabs raised $180 million, indicating strong investor confidence.
- The startup's new model Scribe aims to enhance speech detection capabilities.
- Scribe supports over 99 languages; 25 languages achieve excellent accuracy (less than 5% error rate).
- In benchmark tests, Scribe outperformed Google Gemini 2.0 Flash and Whisper Large V3.
- Scribe features include speaker diarization, word-level timestamps, and auto-tagging of sound events.
- Currently, Scribe can only transcribe pre-recorded audio and a real-time version is in progress.
- Scribe is competitively priced at $0.40 per hour for transcription.
Why it matters
- The emergence of ElevenLabs' speech-to-text model aligns with the growing demand for AI-driven communication tools.
- The AI speech recognition market is expected to grow significantly by 2025, driven by advancements in natural language processing.
- The competition among speech-to-text services promotes innovation in the accuracy and pricing of transcription services.
- The development of real-time transcription capabilities reflects broader IT trends focusing on enhancing remote communication and accessibility.
ElevenLabs' launch of Scribe underlines a pivotal shift in its strategy and highlights the competitive landscape of AI-driven speech recognition, which is poised for significant growth and innovation in the coming years.