Skip to content

Move VAD to akiki.audio and verify both VAD runtimes - #14

Merged
tobyrjackson merged 5 commits into
mainfrom
feat/audio-vad-namespace
Nov 18, 2026
Merged

tobyrjackson merged 5 commits into
mainfrom
feat/audio-vad-namespace

Conversation

@tobyrjackson

Copy link
Copy Markdown
Contributor

Summary

  • Move the Python VAD interface to akiki.audio.vad. The legacy akiki.vad and akiki.VAD imports warn about removal in 0.3.0.
  • Add the vad.Model enum. Akiki-VAD stays the default. TEN-VAD accepts local ONNX weights or an existing pinned Hugging Face snapshot.
  • Run the TEN-VAD network on either the Akiki engine or ONNX Runtime. Both paths use the local Akiki TEN-VAD frontend.
  • Require end-to-end inference for both models on both runtimes in the trusted Linux x86-64 wheel build. The test generates the TEN-VAD ONNX fixture locally and checks frame scores, buffering, streaming state, and reset behavior.

Validation

  • Ruff, YAML lint, and actionlint passed. The offline source suite passed 471 tests and 9 doctests. Five optional MiniLM and private training tests skipped.
  • Built the pinned TEN-VAD library offline. Verified all four model and runtime combinations against a locally installed wheel.

Closes #11.

@tobyrjackson
tobyrjackson merged commit 250f74c into main Nov 18, 2026
4 checks passed
@tobyrjackson
tobyrjackson deleted the feat/audio-vad-namespace branch November 18, 2026 13:07
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.

akiki.vad should be moved into a new top-level akiki.audio

1 participant