bpwhelan/GameSentenceMiner v2026.7.4

Release v2026.7.4

on GitHub

Downloads

Intel Mac builds are no longer provided. If you need GSM on Intel Mac, run it from source.

What's Changed

This release improves capture and OCR reliability, restores TextFeed after reconnects, adds precise audio editing, and makes Anki card review more convenient.

Highlights

1. Anki Confirmation Dialogue - Waveform Enhancements

Click through the waveform to hear certain parts of the audio, Shift + Click to cut out audio from the middle of the audio. (In case there are large sections of silence or sound you want to cut out)

2. TextFeed Session Persistance

When opening textfeed, GSM will add all the lines in the current session that the textfeed doesn't already have. So if you didn't have it open, it will now have all the lines instead of just future lines.

3. Other

• Windows Graphics Capture is more dependable and efficient. Will now capture only the amount of fps that we need for ocr/other operations.
• For Duplicate merging, you can now decide to overwrite instead of merge even if you are using kiku.
• Potential fix for OCR crashing when trying to access old screenshots for deduping.
• Automatically set AnkiBeacon Port on install if using a non-default port
• Black Bar Detection for agressively Pillar-boxed games (i.e. 3DS)

Detailed Changelog

Capture and OCR

• Replaced the legacy Windows capture path with Windows Graphics Capture (WGC), with automatic OBS fallback when direct capture is unavailable.
• Added a configurable frame-rate cap for main-process WGC capture. OCR capture pacing now follows the selected scan rate instead of exposing a second rate setting.
• Added an optional delay before manual green-area OCR scans, with a gamepad-only option for games that need time to redraw after an input.
• Avoided loading unused OCR engines in manual mode and lazily initialized a configured engine when a secondary scan needs it.
• Changed the default static screenshot format to AVIF and added a fast encoder path with a fallback encoder.
• Scaled overlay coordinates against the captured game client so text boxes remain aligned when capture and window sizes differ.

Overlay and TextFeed

• Added Scan on Overlay Activation for Push to Show, controller navigation, and reopening the main text box.
• Migrated the older Push-to-Show rescan preference into the new shared setting and suppressed activation results that are at least 95% similar to the previous scan.
• Restored the complete in-memory TextFeed session in its original order after reconnecting, including active and timed-out lines, without duplicating lines already present.
• Preserved current-session lines that arrive while TextFeed is rebuilding its connection.
• Honored the legacy overlay master gamepad setting before gamepad navigation initializes.

Audio and Card Creation

• Added waveform seeking during stopped and active playback.
• Added removable middle cuts with drag selection, Shift-drag shortcuts, cut undo, and previews that skip the marked ranges.
• Saved edited audio by joining the remaining ranges with short fades.
• Added Overwrite with new context as a one-time choice in the duplicate field-grouping dialog without changing the saved merge preference.
• Enabled Regenerate SentenceMeaning by default in the Anki confirmation dialog.
• Configured downloaded AnkiBeacon add-ons automatically when GSM uses a nondefault port while preserving unrelated webhook settings.

Reliability and Maintenance

• Gave each GSM process its own rotating log files to prevent Windows file-lock conflicts during log rotation.
• Updated the log-opening action and logging documentation for the process-specific filenames.
• Ignored generated Windows audio-capture build artifacts.
• Added regression coverage for capture pacing and fallback, manual OCR delays, overlay rescans and coordinate scaling, TextFeed recovery, waveform edits, Anki workflows, custom-port AnkiBeacon packages, screenshot encoding, and process-isolated logs.

Commit Inventory

• 330755f1 - Add overlay activation rescanning.
• 8fa34b47 - Replace WinAPI capture with configurable WGC.
• 000d1943 - Add waveform seeking and audio cuts.
• 04d978fd - Recover TextFeed sessions after reconnect.
• 3c1f6a74 - Improve manual OCR scan handling.
• 3c497e50 - Default static screenshots to fast AVIF encoding.
• 25b5bb70 - Scale overlay coordinates to captured client size.
• eb3f5baa - Improve Anki confirmation context workflows.
• 2a6373ba - Configure AnkiBeacon packages for custom GSM ports.
• d778039d - Isolate rotating logs by process.
• a1c9e930 - Honor the legacy overlay gamepad master setting.
• 445021a9 - Ignore Windows audio capture build artifacts.
• db8facf7 - Sync the locked project version.
• Release preparation worktree - Make the current-log test portable and align release metadata.

Release Metadata

• Stable baseline: v2026.7.3, published 2026-07-23.
• Repository comparison: v2026.7.3...develop, plus the reviewed release-preparation worktree.
• New-release commit window: v2026.7.3..db8facf7, plus the release-preparation worktree.
• Reviewed head: develop at db8facf7.
• Reviewed scope: 13 commits across 75 branch-diff files, plus the release-preparation test fix and metadata files.