feat(processor): improve adaptive filter tuning with speech metrics and noise-aware thresholds #6
+253
−92
Conversation
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
…ured noise floor Previously, tuneNoiseRemove() used fixed values (-55dB threshold, 6dB expansion) regardless of actual noise characteristics. This meant clean recordings received no adaptive benefit while noisy recordings could have speech unnecessarily affected. Now the compand filter adapts to measured noise from silence regions: - Threshold: noise floor + 5dB (clamped to [-70, -40]) - Expansion: scales with noise severity (4-12dB across 4 tiers) Default values updated to -55dB/6dB as fallback when no noise profile available. Enhanced logging shows measured noise floor and adaptive derivation rationale. Add TestScaleExpansion with 8 test cases covering all expansion tiers. Signed-off-by: Martin Wimpress <[email protected]> feat(processor): prefer speech-specific spectral metrics for adaptive filter tuning Previously, adaptive filter tuning used full-file spectral measurements which are diluted by silence periods. The SpeechProfile struct contains speech-specific versions of spectral metrics that provide more accurate input for tuning filters that process speech content. Now these filters prefer speech-specific measurements when available: - tuneDeesserFull: centroid and rolloff for sibilance detection - tuneLA2ARatio: kurtosis for harmonic structure decisions - tuneLA2ARelease: flux for timing decisions - tuneDC1Declick: centroid for window sizing Falls back to full-file measurements when SpeechProfile is nil or has zero values. Added preferSpeechMetric helper with 4 test cases covering available, zero, negative, and both-zero scenarios. Signed-off-by: Martin Wimpress <[email protected]>
![cubic-dev-ai[bot]](https://avatars.githubusercontent.com/in/1082092?s=60&v=4){kind=small}
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.
Summary
This PR enhances the adaptive filter tuning system to use more accurate measurements for filter parameter selection:
NoiseRemove adaptive thresholding: Replace fixed compand parameters with values derived from measured noise floor during silence analysis. Threshold scales to
noise_floor + 5dB(clamped to[-70dB, -40dB]) and expansion scales from 4-12dB based on noise severity across 4 tiers.Speech-specific spectral metrics: Filters that process speech content now prefer metrics from
SpeechProfile(speech-only regions) over full-file measurements. This provides more accurate tuning fortuneDeesserFull,tuneLA2ARatio,tuneLA2ARelease, andtuneDC1Declick.Fallback handling: Both improvements gracefully fall back to previous behavior when noise profile or speech metrics are unavailable.
Problem Solved
Previous adaptive tuning used full-file spectral measurements diluted by silence periods and fixed noise reduction parameters regardless of actual recording characteristics. This caused:
Changes
internal/processor/adaptive.go: Core adaptive logic withpreferSpeechMetric()helper and updatedtuneNoiseRemove()internal/processor/adaptive_test.go: 12 new test cases for metric preference and expansion scalinginternal/processor/filters.go: Updated default NoiseRemove config valuesinternal/logging/report.go: Enhanced rationale logging for adaptive decisionsTesting
just testpasses with all new and existing testsSummary by cubic
Improves adaptive filter tuning with speech-aware metrics and noise-aware compand thresholds for more transparent, consistent processing. Speech filters now use measurements from speech regions, and noise reduction adjusts to the actual noise floor.
Written for commit 7183b89. Summary will update on new commits.