Repository navigation
Remove deprecated assignment to .shape - #641
Merged
Merged
Conversation
This restores the original behavior of .base pointing directly to the CFFI backend buffer.
thervier
added a commit
to funwithagents/wica-interaction
that referenced
this pull request
Nov 13, 2026
…ells where the echo moved (#14) * Docs: AEC3 all but ignores stream_delay_ms; stop telling users to tune it AEC3 estimates the echo's delay itself and uses an external one only with delay.use_external_delay_estimator (off, and not exposed by the binding): replaying the same audio with the hint at 0 to 500 ms changes the residual by 0.1 dB. The spec no longer says the estimator is "seeded" by it, and the guides no longer say to raise it when the agent hears itself. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * Tests: ignore sounddevice's NumPy 2.5 deprecation, raised in every audio callback sounddevice 0.5.6 assigns `data.shape` in each callback. A plain run ignores the warning, but pytest records each one, ~17k per end-to-end conversation, and processes them at the end of each test: the process froze ~300 ms while the app's audio ran, its mic stream then stayed ~85 ms behind for the rest of the run, and the echo canceller leaked while it relearned (found with a stall probe in CI). Filtered for that message and module only; fixed upstream (spatialaudio/python-sounddevice#641), to drop once a release after 0.5.6 is locked. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * CI: upload the computer's end-to-end evidence on every run Its audio.npz recordings are what the echo canceller is measured on, and the echo check no longer fails, so a failure-only upload left nothing to measure. Kept 30 days. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * e2e: the room holds its echo's delay; the echo path logged as evidence The room recorded the app's speaker on one stream and played it into the cable on another; with their clocks 250-400 ppm apart its echo crept 20-30 ms per conversation, which a laptop's does not and the canceller pays for. It now holds the delay from recording to playback (PortAudio's ADC and DAC times): one resampled sample per block against the drift, the whole excess cut after a late block. Offline, at 0, +-400 and +1000 ppm with a 90 ms late burst, it holds within 0.1 ms (the old room: 50 -> 103 ms over 40 s); in CI the delay seen by the app stays flat. The evidence now also tells where a change in the echo's delay comes from: the room's log of its echo path (room.npz) and each callback of the app's own two audio streams (in audio.npz: latency, underflow and overflow flags). The echo recorder is entered before the build, which opens the voice's output stream. specs/e2e.md: Implemented -> Updated -> Implemented, with its plan Done. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes #638.
I intentionally didn't use
np.reshape(), even though it was suggested in the NumPy warning message. The function call is much slower than the eponymous method call, but both of them change the returned array's.baseattribute, which then points to a temporary array. I would like it to, as it has done before, point directly to the CFFI backend buffer.