Quicksubs Integration

On-Device Transcription & Audio Alignment (quicksubs Integration)

RightSub integrates with quicksubs, a high-performance on-device transcription CLI tool developed by Matt Birchler.

• Author: Matt Birchler
• GitHub Repository: mattbirchler/quicksubs
• Author's Site: Birchtree
• Mac App Companion: Quick Subtitles

---

Overview

RightSub provides native on-device speech-to-subtitle and audio-guided retiming powered by quicksubs:

1. Ingestion Fallback (Use Case 1): Transcribes raw video files into .srt subtitles when no embedded or external subtitles exist.
2. Audio-Guided Alignment (Use Case 2): Re-synchronizes drifted or misaligned subtitles by extracting ground-truth speech timestamps directly from the audio track.

---

Installation

brew install mattbirchler/tap/quicksubs

---

Supported Speech Engines

Engine	Flag	Model Size	Description
Apple SpeechAnalyzer	--engine apple	0 MB (Built-in)	On-device Apple Silicon Neural Engine execution. Default.
OpenAI Whisper	--engine whisper	~626 MB	Highest transcription accuracy for challenging audio.
NVIDIA Parakeet	--engine parakeet	~400 MB	Fast, lightweight local neural model.

---

Commands & Usage

1. Direct Transcription

python3 scripts/00_transcribe_audio.py "/path/to/video.mp4" -e apple

2. Extraction with Automatic Fallback

python3 scripts/01_extract_subtitles.py "/path/to/videos" --transcribe --engine apple

3. Audio-Guided Synchronization

python3 scripts/16_audio_align_sync.py "unsynced.he.srt" -o "aligned.he.srt" -v "video.mp4"

---

Future Roadmap

• Zero-Bandwidth Cloud Optimization: Local STT, sending only lightweight text cues to Gemini.
• Media Server Folder Actions: macOS LaunchAgent for automated drop-folder captioning.
• Local Podcast Workflows: Direct transcription of audio memos and lectures.
• Apple Silicon Profiling: Benchmarking local models via quicksubs bench.