Reverb ASR+Diarization: Best open-source ASR for long-form audio

Reverb, an open-source automatic speech recognition system with integrated speaker diarization optimized for long-form audio, was posted on HackerNews as a Show HN with 12 points. This provides a free alternative to commercial ASR services for transcription and meeting analysis, particularly relevant for agent-based voice workflows.

For builders of voice agents and meeting analysis tools, Reverb removes the per-minute cost of cloud ASR APIs and enables on-premise deployment, reducing both latency and privacy exposure. The integrated diarization eliminates the need to chain separate speaker identification services, simplifying pipeline architecture. A second-order effect is increased feasibility of running real-time transcription for long, multi-speaker conversations in edge or air-gapped environments, which previously required expensive or closed-source solutions. This shifts the cost baseline for voice agents from variable API fees to fixed compute overhead.