Skip to content

Add on-device speech capacity and quality evidence - #20

Open
haaziq-sl wants to merge 7 commits into
mainfrom
codex/android-speech-capacity-20261118
Open

haaziq-sl wants to merge 7 commits into
mainfrom
codex/android-speech-capacity-20261118

Conversation

@haaziq-sl

@haaziq-sl haaziq-sl commented Nov 18, 2026 •

Copy link
Copy Markdown
Collaborator

Summary

  • Add a phone speech intake and a native timing import for open-loop concurrency sweeps. Evidence includes first PCM, simulated first audio, per-stream p95 real-time factor, underruns and throughput. First audio can be tracked without a latency gate.
  • Accept native PCM/WAV for speech quality checks and support an independent A/A control for exact output. Keep valid scorer observations when UTMOSv2 returns an out-of-range window.
  • Expose warmup fixtures, first-PCM comparisons and speech ledger fields. Add optional Nsight Systems capture range end handling and document wheel installation on restricted hosts.
  • Add opt-in maximum-performance and shared-device efficiency policies for heterogeneous inference across phones, laptops and desktops.

Refs #12, #13, #15, #16, #17, #18 and #19. The PR addresses the clauses above.

Validation

  • The full local suite passed with 1,984 tests and 90 skips. The final older-client compatibility follow-up passed its 31 audio and media-round tests.
  • The local wheel build, hash-pinned install, and MCP stdio checks passed.
  • Intake and skill checks for the resource policies passed.
  • A native Android timing capture exercised the import.

@haaziq-sl
haaziq-sl marked this pull request as ready for review November 18, 2026 10:06
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant