sttlab-tech
8bd1a6caf8
SwiftUI app, the one-click way to use the local transcription pipeline — installs and runs with no separate manual steps. The Xcode build phase (scripts/embed-binaries.sh) builds audiotee (git submodule, pinned to a commit on our own Gitea fork) and transcriptor-ai + qwen_asr (sibling checkout, which itself vendors qwen-asr as a submodule), then embeds all three binaries in the app bundle. The AI model is the one thing still fetched at runtime, via ModelManager.swift, since it's multi-GB and doesn't belong baked into a build. Also implements: model download with real progress UI, capture source selection (system/mic/both), a Settings window for the transcript save location and a glossary file (biases ASR toward proper nouns/product names via qwen_asr's --prompt), and a live chat-bubble transcript view. See CLAUDE.md for the full architecture, the reasoning behind each decision, and a fairly long list of real bugs found via actual testing (not just code review) — TCC permission escalation through a subprocess tree, SwiftUI Form/Grid layout quirks, @State vs @AppStorage persistence, a prompt-leakage bug in the glossary feature, and more. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
13 lines
300 B
Plaintext
13 lines
300 B
Plaintext
.DS_Store
|
|
xcuserdata/
|
|
*.xcuserstate
|
|
DerivedData/
|
|
/build/
|
|
*.dSYM
|
|
*.dSYM.zip
|
|
|
|
# Manual ASR quality test fixtures (downloaded video + subtitles used to
|
|
# compute WER against transcriptor-ai) — large binary media, not source.
|
|
# See CLAUDE.md for what's in here and the results obtained from it.
|
|
/test/
|