Redefining What’s Possible in Open-Source ASR and Diarization

In a groundbreaking move that promises to reshape the landscape of speech recognition technology, Rev is proud to announce the open-sourcing of Reverb, our cutting-edge Automatic Speech Recognition (ASR) and diarization models.

This bold step forward represents not just a technological achievement, but a philosophical shift in how we approach AI development and accessibility in the voice technology space.

Why Reverb Matters: A Game-Changer for the Industry

Reverb isn’t just another ASR model — it’s a paradigm shift. Trained on an unprecedented extreme-quality 200,000 hours of human-transcribed English speech, Reverb sets a new benchmark for accuracy and versatility in speech recognition. But what truly sets Reverb apart is its commitment to openness and flexibility.

1. Unparalleled Accuracy: Reverb outperforms all existing open-source speech recognition models across various long-form speech recognition domains. This isn’t just an incremental improvement; it’s a leap forward in what’s possible with open-source ASR.
2. Democratizing Advanced AI: By open-sourcing Reverb, we’re putting enterprise-grade speech recognition capabilities into the hands of researchers, developers, and innovators worldwide. This move has the potential to accelerate advancements in voice technology across countless fields.
3. Verbatimicity Control: Unique to Reverb is the ability to control the level of verbatimicity in transcriptions. This feature opens up new possibilities for applications ranging from precise legal transcriptions to more readable content for captioning.
4. Diarization Excellence: Reverb doesn’t just transcribe; it understands who’s speaking. Our state-of-the-art diarization models, fine-tuned on 26,000 hours of expertly labeled data, set a new standard for speaker attribution in multi-speaker environments.
5. Flexibility for Researchers and Developers: We’re releasing both full production pipelines for developers and pared-down research models for experimentation. This dual approach ensures that Reverb can drive innovation in both academic and commercial settings.

Introducing Our New ASR Models and Flexible Licensing

As part of this revolutionary release, we’re thrilled to introduce two new ASR models that push the boundaries of what’s possible in speech recognition:

1. Reverb V1: Our flagship model, offering unparalleled accuracy at just 20 cents per hour.
2. Reverb Turbo V1: A lightning-fast option that maintains high accuracy at an incredibly competitive 10 cents per hour.

Both models include state-of-the-art diarization capabilities, providing a complete solution for transcription and speaker attribution.

For those leveraging our open-source models, we’re offering highly competitive pricing:

• Reverb Self-Hosted: 20 cents per hour
• Reverb Turbo Self-Hosted: 10 cents per hour

We understand that different projects have different needs, which is why we’re also offering “all you can eat” licensing options. Reach out to us to discuss how we can tailor our offerings to your specific requirements.

Empowering Research and Innovation

We believe in the power of open collaboration to drive innovation. That’s why we’re introducing the “Rev Model Non-Production License,” which allows anyone to use our models for evaluation, research, and personal use at no cost. This license opens up a world of possibilities for academics, hobbyists, and innovators to explore and build upon our technology without financial barriers.

For commercial applications, we offer a proprietary license. We’re committed to working with businesses of all sizes to find the right licensing solution that enables them to leverage the power of Reverb in their products and services. For information on usage-based or all-inclusive commercial licenses, please contact us at licensing@rev.com.