Skip to content

AI & Automation1 min read

ElevenLabs releases Scribe v3 transcription models

New speech-to-text updates from ElevenLabs aim to improve accuracy for real-time and recorded audio applications.

by Fini MironFounder @ Balkaris
An abstract composition printed flat in pale cool grey, steel blue, deep navy and bright mustard yellow: four large rectangles lying flat and overlapping like sheets of paper on a table, each offset slightly to the right of the one beneath it.
Drawn for this article on our own machine. It is not a photograph.

ElevenLabs has launched two new speech-to-text models, Scribe v3 and Scribe v3 Live, according to a report by The-decoder. These updates aim to improve transcription accuracy for both pre-recorded audio and real-time voice applications. The company claims the new models handle background noise and overlapping speakers more effectively. Scribe v3 Live specifically uses conversational context to prevent voice agents from interrupting users mid-sentence. These tools are currently available through the ElevenAPI and ElevenAgents platforms.

Why it matters

For businesses integrating voice AI, the trade-off between accuracy and cost is now more defined. While Scribe v3 offers high precision for complex audio, it is not the cheapest or fastest option on the market. You must decide if the reduction in word errors justifies the higher price point compared to providers like Microsoft or xAI. If your use case involves high-volume, simple transcriptions, these new models may be an unnecessary expense. However, for specific applications, the benefits include:

  • Improved user experience in voice agents by reducing accidental interruptions.
  • Better handling of specialised vocabulary like brand names or medical terms.
  • Enhanced data privacy for EU-based operations through zero-retention modes.

What to watch

You should test how these models handle your specific industry vocabulary. ElevenLabs allows you to preload specialised terms like brand names to improve accuracy. If you operate in a highly regulated sector, check if the zero-retention mode and EU-based servers meet your data compliance requirements. Accuracy is useful, but only if the model recognises your specific domain terms correctly.

Within a year, Scribe models will become the standard for high-fidelity audio processing. This happens only if ElevenLabs reduces the per-minute cost to match current transcription benchmarks.

Key takeaways

  • Higher transcription accuracy often requires a higher price point.
  • Real-time models use conversational context to manage speaker interruptions.
  • Specialised vocabulary requires manual preloading to ensure correct recognition.

Frequently asked

How do these models compare to existing transcription services in terms of cost?

They are likely more expensive than providers like Microsoft or xAI. You must weigh the cost against the reduction in word errors for your specific use case.

Can these models be used for sensitive data in Europe?

Yes, if you use the zero-retention mode and EU-based servers. You should verify these settings meet your specific regulatory requirements.