ASRSubs

ASRSubs is a desktop app for generating subtitles from local media files without sending the audio to a hosted transcription service. You pick a file, choose a local model, let the app transcribe and align the speech, then review and save the resulting .srt file.

It is built with Wails and ships with a managed Python runtime plus bundled media tools so the app can prepare audio, run local ASR models, and generate subtitle timing on the machine itself.

What The App Does

• runs speech-to-subtitle workflows locally
• supports both audio and video inputs
• downloads transcription models on demand and reuses them later
• prepares media automatically before transcription
• shows stage-by-stage progress during runtime setup, downloads, transcription, alignment, and subtitle building
• opens the finished subtitle draft in an editor so you can make final fixes before saving

Typical Workflow

1. Open the app and choose a media file.
2. Prepare the local runtime on the first run.
3. Download the transcription model you want to use.
4. Click Start Transcription.
5. Review the generated subtitle draft.
6. Save the final .srt file where you want.

Supported Media

ASRSubs currently accepts:

• .wav, .mp3, .m4a, .aac, .flac, .ogg, .opus, .mp4, .mov, .m4v, .mkv, .avi, .webm

Available Models

The app currently exposes two local ASR model options:

• Qwen3-ASR-1.7B Best accuracy, slower, larger first download.
• Qwen3-ASR-0.6B Faster startup, lighter footprint, good for smaller machines and quicker runs.

The app also manages an internal forced aligner automatically when timestamp generation needs it.

What You Can Adjust

Inside the settings drawer you can:

• switch between the available local models
• download or remove model files
• change subtitle grouping preferences
• change max line length and lines per subtitle
• change the alignment chunk size used for longer files
• re-run local runtime preparation if the managed runtime needs repair

Project Status

The repository currently builds desktop packages for:

• macOS
• Windows

Windows builds are published as both a portable ZIP and an installer. macOS builds are packaged as a DMG.

Development

Run the app in development mode:

wails dev

Build the frontend assets directly:

cd frontend && npm run build

Packaging

macOS

Build the packaged macOS app and DMG:

./scripts/build-macos-package.sh

That flow:

• builds ASRSubs.app
• stages the managed runtime, worker script, requirements, ffmpeg, and ffprobe
• creates build/bin/ASRSubs.dmg

Important staging inputs:

• ASRSUBS_PYTHON_STANDALONE
• ASRSUBS_FFMPEG_PATH
• ASRSUBS_FFPROBE_PATH

If those are not set, the staging helper falls back to packaged runtime/tool locations in the repo or the current shell PATH.

Windows

The repository produces:

• ASRSubs-<version>-windows-amd64-portable.zip
• ASRSubs-<version>-windows-amd64-installer.exe

The Windows GitHub Actions workflow:

1. installs Go, Node, and Python toolchains
2. installs ffmpeg and NSIS
3. builds ASRSubs.exe
4. stages the bundled runtime and tools
5. creates the portable ZIP and installer

Verification Helpers

./scripts/verify-macos-package.sh
./scripts/verify-windows-package-layout.sh
./scripts/verify-windows-workflow.sh

Notes For Unsigned Builds

• macOS may show a Gatekeeper warning the first time you open a local unsigned build
• Windows may show a SmartScreen warning for unsigned artifacts

For local testing, that is expected.