Local ASR decode median

Local speech recognition decodes in about 170 ms, far faster than the cloud round trip it replaced.

Single reading

167.2 ms

MEASURED recorded 6 August 2026

Protocol

Test definition

probe_conversation_latency: local GPU ASR (float16) decode of utterances, compared with a live provider round trip.

Limits

Limitations

  • Sample count not recorded; provider comparison (2479 ms) from a separate live measurement; end-to-end first-answer audio explicitly UNMEASURED in this probe.

Log

Recorded values

Date Value Evidence Samples Source
6 August 2026 167.2 ms MEASURED evidence/CONV/probe_conversation_latency.json

Source paths refer to the private JARVIS repository and runtime records. They are listed for audit and are not published.