Home Projects About GitHub Repository
v0.1.0 Active Release GPL-3.0 Python 3.10+ PySide6 (Qt) Qwen3-ASR (1.7B) PyTorch CUDA / Metal (MPS) FFmpeg VAD Pacing SRT / WebVTT

SubtitleGo

A standalone, high-performance desktop application for automatic speech recognition, smart silence-based pause detection, synchronized video playback, and natural-paced subtitle generation.

View Source on GitHub Release Notes

Key Features

Direct In-Process Inference

Runs Qwen3-ASR (1.7B) locally inside process memory without external web servers, background REST daemons, or Docker containers. Automatic GPU acceleration with CPU fallback.

Smart Pacing Engine

Intelligent speech pause segmentation, clause splitting, line character limits (42 Latin / 18 CJK), and timestamp interpolation targeting 2.0s to 4.5s per cue.

52 Languages & Dialects

Multi-language speech recognition with automatic language identification, dialect adaptation, and CJK line wrapping rules.

Video Preview & Subtitle Overlay

Built-in multimedia player with live synchronized high-contrast subtitle overlay, click-to-seek, and active cue auto-scroll during playback.

Interactive Cue Studio

Live subtitle text search with instant highlight, in-place text editing with real-time video sync, and one-click timestamp adjustments.

Multi-Format Synchronization

Bidirectional synchronizer supporting Cue Table, SRT View, and WebVTT View with one-click clipboard copy and auto-saving alongside source media.

Hardware Compatibility

Platform / Component Tested Environments Acceleration Mode
Operating System Tested: Windows 11 (64-bit)
Tested: Linux (Ubuntu, Debian, Fedora, Arch)
Tested: macOS Sonoma / Sequoia
Native desktop binary with bundled FFmpeg multimedia tools.
GPU Acceleration Tested: NVIDIA RTX 50-Series (Blackwell)
Tested: Apple Silicon Metal (M1 - M4 Series)
NVIDIA RTX 40 / 30 / 20 Series
FP16/BF16 tensor acceleration via CUDA (Windows/Linux) and Metal Performance Shaders (macOS).
CPU Fallback Intel / AMD x64 processors
Apple Silicon CPU
Automatic CPU execution fallback when no compatible GPU is detected.

Getting Started

1

Pre-built Executable

Download SubtitleGo-v0.1.0-windows-x64.zip from GitHub Releases, extract to any folder, and run SubtitleStudio.exe. No Python setup required.

2

Build from Source

For developers looking to inspect the codebase, customize inference pipelines, or build on Linux/macOS, complete setup guides and requirements are available in the repository.

View Source on GitHub

Frequently Asked Questions & Technical Specs

Does SubtitleGo require an internet connection or cloud API key?

No. SubtitleGo executes the 1.7-billion parameter Qwen3-ASR model locally in computer memory. All speech recognition, voice activity detection, and subtitle timestamping operate completely offline with zero telemetry or cloud server communication.

What GPU hardware acceleration is supported?

SubtitleGo supports NVIDIA CUDA (RTX 50-series Blackwell, RTX 40/30/20 series) on Windows and Linux, Apple Silicon Metal (MPS on M1-M4) on macOS, and automatic multi-threaded CPU fallback for systems without dedicated tensor hardware.

Which subtitle formats are supported for export?

SubtitleGo exports standard SubRip (.srt) and WebVTT (.vtt) files. The editor includes real-time bidirectional synchronization between the visual cue table and raw text views with one-click clipboard copying.

How does the smart pacing engine segment speech?

The pacing engine utilizes Voice Activity Detection (VAD) to split sentences at natural silence pauses, enforces line length limits (42 Latin or 18 CJK characters), and bounds each subtitle duration between 2.0s and 4.5s for optimal viewer reading speed.