Repository navigation
berd-call: bring Spokesperson playback closer to Siri loudness - #411
Closed
johnmatthewtennant wants to merge 1 commit into
Closed
johnmatthewtennant wants to merge 1 commit into
johnmatthewtennant wants to merge 1 commit into
Conversation
Signed-off-by: Nora <49aa1f65411fd096d2e2ec144f1e7aa36fdc76d1b907cfdf7be000c66f9d3b8e@buzz.block.builderlab.xyz>
Contributor
Author
|
🤖 Folded into #416 for review and merge. |
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Spokesperson replies are quieter than Siri agent replies. Capturing the same two sentences from Siri Aaron and Realtime Marin at normal speed measured active RMS of −17.9/−18.9 dBFS versus −22.1/−22.1 dBFS, before system volume or speakers.
Apply up to 1.6× gain (4.1 dB) to the Spokesperson's playback stream, with an immediate sample-peak limiter at 0.95 and 250 ms recovery. The captured Marin samples then measure −18.1/−18.1 dBFS. Siri playback is unchanged. Silence remains silent and non-finite samples become zero.
Related issue
none found
Testing
Quiet-speech and limiter regressions fail with the original pass-through and pass with compensation. A playback-host regression checks the samples reaching the real output interface, including unchanged frame counts. Coverage also includes loud and clipped input, silence, non-finite samples, limiter recovery, and chunk boundaries.
The captured PCM was processed through the production Rust compensator. On the first sentence, integrated loudness moved from −23.0 to −19.0 LUFS, compared with Siri's −18.8 LUFS. Its processed true peak measured −0.4 dBTP. These two samples calibrate the proposed boost; John's listening comparison remains pending on the combined test build.
Listen to the attached comparison at one fixed device volume: Siri Aaron, original Realtime Marin, then the same Marin PCM processed by this PR’s production Rust module. The clip uses the sentence “The call is ready. I can help you review the changes and decide what to do next.” The original capture has no post-processing; the compensated segment adds only this PR’s gain and limiter. This demonstrates PCM compensation, while live-call device playback and John's final listening judgment remain pending.
comparison.mp4