Audio
Open ASR transcribes. EdgeLance attributes.
Open-source acoustic models do the transcription. EdgeLance adds the speaker, the doctrine, the citation, and the replay.
Open ASR
Acoustic model.
Open-source transcription baseline.
- Open-weight acoustic models (multilingual)
- Pre-trained on web-scale audio
- Word tokens with timestamps
- Permissive licensing (MIT, Apache 2.0)
- CPU, GPU, and NPU inference
- Quantized variants for mobile
- WhisperKit and Distil-Whisper builds
- ASR baseline of record
EdgeLance Audio
Voice intelligence.
Doctrine-grade comprehension.
- Source-profile VAD (radio, body-worn, camera, clean)
- Voice fingerprint speaker ID against enrolled bank
- Doctrine normalizer (brevity, callsigns, hallucination filter)
- Citation anchor: every token bound to source-media timestamp
- Continuous surveillance with threat-phrase triggers
- Voiceprint threat watch list with severity routing
- Doctrine compliance scoring with reviewer feedback loop
- Closed-loop adapter: squad-bound LoRA under burn lifecycle
Backed by provisional patent applications filed with USPTO in June 2026 covering capacity-governed multi-net radio comprehension with joint diarization-callsign attribution, doctrine-compliance scoring with voice-fingerprint attribution, and closed-loop edge model adaptation with operator-anchored corrections.