ElevenLabs launches its letter to the text

elevenThe start of the artificial intelligence that has just raised A $ 180 million financing tourHe was primarily known for the brilliant generation of sound. The company took another technological direction by launching the first speech model to the independent text called SCRIBE.

Startup, Worth $ 3.3 billionSeveral other companies have helped provide speech services to the text through their wide votes library. However, the company is now looking to enter into disclosure and competition with proverbs swordand Speakingand Associationand DeepgramAnd Openai’s Whisper Models.

The writer ElevenLabs model supports more than 99 languages ​​at all. The company classifies more than 25 languages ​​in an excellent accuracy category for the model where the word error rate is less than 5 %. This menu includes the English language (97 % accuracy rate), French, German, Indian, Indonesian, Japanese, Kanada, Malayam, Polish, Portuguese, Spanish, and Vietnamese. Other languages ​​are classified in different categories with an error rate in words 5-10 %), word error rates from 10 to 20 %), and word error rates (25 to 50 %).

The company said the model is outperforming Google Gemini 2.0 Flash and Homper V3 through multiple languages ​​in Flores tests and common audio tests.

ElevenLabs developed the speech component to the text of the AI ​​Agent Agent platform, which was released last year. However, this is the first time The company issues a form to detect independent speech. In a conversation with Techcrunch last month, CEO Mati Staniszewski spoke about improving speech detection models.

“We want to understand what you are in a better conversation. Staniszoski said at that time:“ We are working on ways to stay away from generating content and understanding and lacking only. ”Many people say that talking to the text represents a problem of solutions. But for many languages, it is It is very bad.

The model also contains a smart speaker to tell you who speaks, and the time levels of words to get accurate translations, and automatic signs of sound like the audience’s laugh. The startup provides a way to customers to copy the video content directly to add translations or clarification in its studio.

SCRIBE only currently works with pre -recorded sound formats. The company said it would issue a low version in the actual time of the form soon. This means that it is not yet effective to meet the copies or write down vocal notes.

ElevenLabs is the author’s pricing at $ 0.40 for an hour of written sound. While the rate is competitive, Some of its competitors Provide a lower price For audio copies at the present time with some distinction features.