Reverb: Open-Source ASR and Diarization from Rev

TL;DR

Reverb delivers open-source, non-commercial ASR and diarization designed for long-form speech, built on a large, human-transcribed English corpus. The ASR component uses a WeNet-based, conformer-CTC/attention architecture with verbatimicity control and a production WFST-based decoder, plus a Turbo variant for speed/memory efficiency. Diarization employs pyannote.audio with a Rev-finetuned v1 and a WavLM-based v2 to improve speaker attribution, evaluated via Word Diarization Error Rate on multiple long-form benchmarks. Together, the releases provide a practical, configurable foundation for researchers and developers to benchmark, tune verbatimity, and deploy end-to-end diarized transcripts in real-world settings.

Abstract

Today, we are open-sourcing our core speech recognition and diarization models for non-commercial use. We are releasing both a full production pipeline for developers as well as pared-down research models for experimentation. Rev hopes that these releases will spur research and innovation in the fast-moving domain of voice technology. The speech recognition models released today outperform all existing open source speech recognition models across a variety of long-form speech recognition domains.