Reverb ASR+Diarization: Open Source Long-Form Audio ASR

Reverb ASR+Diarization, an open-source speech recognition system combining automatic speech recognition with speaker identification, gained traction on HackerNews, indicating builder interest in on-premise audio processing.

For voice AI applications, this reduces reliance on cloud ASR APIs (Deepgram, AssemblyAI, Google Cloud Speech) where per-minute costs accumulate at scale. Diarization—identifying who spoke when—is typically a separate service layer; bundling it locally eliminates round-trip latency and per-request billing. This matters operationally for call centers, meeting transcription platforms, and compliance-heavy workflows where audio residency requirements or volume economics demand local inference.

Builders can now deploy end-to-end voice pipelines without multi-vendor dependencies. This shifts infrastructure from API-first architectures toward GPU-capable self-hosted systems. Second-order: teams previously blocked by API costs or data residency constraints can now prototype voice features. However, this trades cloud convenience for operational burden—model optimization, hardware provisioning, and maintenance fall to the operator rather than the vendor.