feat(processor): add voice-activated detection and adaptive speech RMS minimum

[flexiondotorg]

Summary

Add automatic detection of voice-activated recordings (Riverside/Zencastr) and adaptive speech RMS minimum threshold. Voice-activated mode detects when recordings are triggered by voice activity (significant silence fraction), automatically adjusting tolerance and processing parameters. Speech RMS minimum adapts from fixed -40 dBFS to a formula based on spectral centroid, improving filter adaptation for diverse recording conditions.

Changes

• Detect voice-activated recordings from digital silence fraction in candidate selection
• Widen speech interruption tolerance for voice-activated mode (2.0-3.0s instead of fixed)
• Disable NoiseRemove compand expansion for uncalibrated recordings to avoid over-expansion
• Implement adaptive speech RMS minimum (speechRMSMinimum = max(-70, cumulativeSum * 0.8 - 22))
• Add voice-activated indicator and compander annotation to analysis display
• Update documentation for detection logic and silence candidate filtering
• Update README with voice-activated recording capabilities
• Improve test coverage for silence detection and analysis display output

Testing

• Unit tests verify silence fraction detection for voice-activated recordings
• Analysis display tests confirm correct output formatting with voice-activated and compander annotations
• Adaptive RMS formula tested with diverse speech profile measurements
• Anna's recording now detects 6 silence candidates instead of 1 after enabling voice-activated detection

[cubic-dev-ai]

5 issues found across 13 files

Confidence score: 3/5

• There is concrete regression risk in internal/processor/adaptive.go: if NoiseRemoveCompandEnabled is not re-enabled on the valid-profile path, one uncalibrated file can keep compand disabled for subsequent files in the same run.
• internal/processor/analyzer_candidates.go has user-facing detection risk: voice-activated mode is only inferred from silence candidates >=8s, so shorter digital-silence gaps may never activate widened speech tolerance.
• Logging/reporting issues in internal/logging/analysis_display.go and internal/logging/report.go can hide or misstate recording mode/disable reasons, which reduces operator visibility and can make troubleshooting harder even when processing succeeds.
• Pay close attention to internal/processor/adaptive.go, internal/processor/analyzer_candidates.go, internal/logging/analysis_display.go, and internal/logging/report.go - processing-state carryover and mode/disable-status reporting are the main risk areas.

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="internal/processor/adaptive.go">

<violation number="1" location="internal/processor/adaptive.go:689">
P1: Re-enable `NoiseRemoveCompandEnabled` on the valid-profile path; otherwise one uncalibrated file disables compand for every later file in the same run.</violation>
</file>

<file name="internal/logging/analysis_display.go">

<violation number="1" location="internal/logging/analysis_display.go:130">
P2: The voice-activated indicator is omitted whenever an elected silence candidate exists, so analysis output hides the detected recording mode in the common success path.</violation>

<violation number="2" location="internal/logging/analysis_display.go:255">
P2: This message hard-codes the wrong disable reason; the compander is also disabled for invalid noise-floor calibration, not just a missing noise profile.</violation>
</file>

<file name="internal/logging/report.go">

<violation number="1" location="internal/logging/report.go:1613">
P2: The voice-activated indicator is only printed in some silence-report branches, so it disappears for reports that have a `NoiseProfile` or only `SilenceRegions`.</violation>
</file>

<file name="internal/processor/analyzer_candidates.go">

<violation number="1" location="internal/processor/analyzer_candidates.go:1030">
P1: This only detects voice-activated mode from >=8s silence candidates, so recordings with shorter digital-silence gaps never enable the widened speech tolerance.</violation>
</file>

_{Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.}

[cubic-dev-ai]

1 issue found across 2 files (changes from recent commits).

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="internal/processor/analyzer.go">

<violation number="1" location="internal/processor/analyzer.go:207">
P2: `NoiseFloorSource` is not updated when elected silence overrides `NoiseFloor`, so the exported source can disagree with the actual noise-floor value.</violation>
</file>

_{Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.}

[cubic-dev-ai]

0 issues found across 5 files (changes from recent commits).

_{Requires human review: Modifies core audio processing logic, including adaptive thresholding and filter chain configuration, which requires human domain expertise to validate.}