2.9 KiB
Transcript Forensics · product overview
Pitch
Audio counterpart to Blocao video forensics. Recordings of meetings, phone calls, depositions are processed on-site by Whisper + pyannote. The console offers natural-language query ("what was said about contract Acme and who was most insistent") with speaker identification, topic detection, and waveform-anchored playback.
Built on the same platform as Blocao: same router, same Cell hardware (with a microphone or audio ingest interface), same MQTT bus, same GitOps, same hub.
What you get
- Transcript console with chat-style query, waveform with speaker bands, transcript navigable by speaker, audio player with live caption.
- Speaker library: voiceprints stored per-customer, automatically matched against new sessions.
- Topic detection: NER + clustering surfaces what was discussed; per-speaker breakdown.
- Insistence ranking: identifies who pushed hardest on each topic (intervention rate, repetition).
- Pin to case: same case management as Blocao video.
- Cross-product correlation (Year 2): correlate audio transcripts with video events at a site.
Use cases
- Meeting forensics for legal/compliance: review of depositions, board meetings, contract negotiations.
- Customer service quality: search support call recordings for problematic patterns.
- Mediation and arbitration: who said what, when, with audio proof.
- Insurance claim review: verify what was promised in pre-purchase calls.
- Internal investigations: HR or compliance review of recorded conversations.
Why same platform as Blocao
The transcriptor and Blocao share enough infrastructure that running them on separate stacks would be wasteful:
- Same router, same Cell, same MQTT, same GitOps, same hub.
- Same console patterns (chat, waveform/timeline, player, case management).
- Same sovereignty model.
- Same evidence chain (post-MVP).
The only product-specific pieces are:
- The Whisper + pyannote container in the Cell stack.
- The speaker library schema in Qdrant.
- The transcript-specific console panel.
See shared-stack.md for the breakdown of shared vs distinct.
Differentiators
| Vs | Transcript Forensics differentiator |
|---|---|
| Otter.ai / Sonix / Fireflies | On-site processing + sovereignty + speaker library across sessions |
| Microsoft Azure Speech / AWS Transcribe | No vendor lock-in + open-source models + same UX as Blocao |
| Open-source Whisper alone | Fleet-managed + speaker identification + topic detection + UX |
Status
Mockup complete (mockups/transcript-forensics.html). Backlog at backlog/transcript-sprint-backlog.md. Many stories shared with Blocao backlog (see backlog/shared-stories.md).
The transcript product launches after Blocao MVP is shipping (target: Year 2). Until then, design lives in this repo to ensure the platform is built with audio in mind from day one.