ElevenLabs launches 'Scribe': stand-alone speech-to-text model for global languages

AI startup ElevenLabs, known for its cutting-edge audio-generation technology, has unveiled its first stand-alone speech-to-text model, named Scribe.

The company, fresh off a $180 million funding round and valued at $3.3 billion, is now expanding its technology portfolio to compete in the speech detection arena.

Scribe supports over 99 languages at launch, with more than 25 languages achieving an “excellent” accuracy rating—defined as a word error rate of less than 5%. This list includes English, with a claimed accuracy rate of 97%, as well as French, German, Hindi, Indonesian, Japanese, Kannada, Malayalam, Polish, Portuguese, Spanish, and Vietnamese. Other languages are categorized into high, good, and moderate accuracy levels based on their word error rates.

According to benchmark tests using FLEURS and Common Voice datasets, Scribe has outperformed competitors such as Google Gemini 2.0 Flash and OpenAI’s Whisper Large V3 across multiple languages. Previously, ElevenLabs developed a speech-to-text component for its AI conversational agent platform, but Scribe marks the first time the company is releasing a dedicated, stand-alone speech detection model.

CEO Mati Staniszewski told TechCrunch last month, “We want to understand what’s being said by you in a conversation better. We are working on ways to move away from only generating content and understanding and transcribing speech.” He noted that while many consider speech-to-text a solved problem, performance for many languages remains suboptimal. “We think we can build better speech detection models because we have in-house teams to annotate data and give us quick feedback,” Staniszewski added.

In addition to accurate transcription, Scribe incorporates smart speaker diarization to identify who is speaking, provides word-level timestamps for precise subtitle generation, and auto-tags sound events such as audience laughter. The model currently processes pre-recorded audio formats, and ElevenLabs plans to release a low-latency real-time version in the near future, which would extend its use to meeting transcriptions and live voice note-taking.

Scribe is priced competitively at $0.40 per hour of transcribed audio, although some rival services offer lower prices with different feature sets. As ElevenLabs continues to push the boundaries of generative AI technology, the launch of Scribe marks another significant step in expanding its influence across both audio-generation and speech detection markets.

Copied