Subtitle LLM: A Subtitle Translation Tool Based on Large Language Models

Subtitle LLM is a command-line tool developed by al123xiaaaa and released on GitHub on June 4, 2026. It uses large language models to translate .srt subtitle files. Its core advantages include support for bilingual subtitle generation, context-aware translation, and automatic translation quality diagnosis, aiming to solve pain points in professional subtitle translation. Original project link: https://github.com/al123xiaaaa/subtitle_llm

Subtitle translation differs significantly from general text translation. It needs to handle colloquial expressions, technical terms, and cultural differences while maintaining timeline synchronization. Traditional machine translation struggles to accurately address these scenarios, and manual translation is time-consuming and labor-intensive. The subtitle_llm project attempts to leverage the context understanding capabilities of large language models to provide a more intelligent subtitle translation solution.

Subtitle LLM supports multiple input formats: SRT subtitle files, word-level transcription JSON files, and video URLs. Its core features include:

1. Context-aware translation to ensure terminology consistency;
2. Parallel processing of subtitle blocks to improve efficiency;
3. Preservation of subtitle timelines and formats;
4. Built-in quality assurance mechanism;
5. Support for multiple LLM providers and custom configurations.

The translation process is divided into four stages:

1. Context Generation: Generate video topic, subtitle summary, and list of untranslatable terms;
2. Block Processing: Split subtitles into blocks for parallel processing;
3. Iterative Refinement: Initial translation → Refinement with context → Quality diagnosis → Re-translate suspicious blocks or review via TUI;
4. Merge and Output: Generate bilingual .srt files.

The quality diagnosis mechanism includes two types:

• Observable Failure Detection: Identify issues such as missing lines, residual placeholders, abnormal length, duplicate translations, and language mismatches;
• Intelligent Diagnosis Aggregation: Generate block-level reports for repeated or cascading failures to efficiently guide re-translation, instead of listing errors line by line.

Installation and Configuration

• Installation: pip install -e .
• Set API key: export GEMINI_API_KEY=\"...\"
• Translation command: subtitle-llm translate --input input.srt --target-language Chinese
• Supports custom configuration files, which can override default LLM providers and model settings

Other Commands

• Download video: subtitle-llm download \"video URL\" --output-dir path --source-language en
• Transcribe audio: subtitle-llm transcribe audio_file --output output_path --language English

Application Scenarios

Suitable for scenarios such as educational video translation, technical lecture subtitle processing, multilingual content creation, and personal learning material translation.

Summary

Subtitle LLM demonstrates the application potential of large language models in professional translation scenarios. Through multi-stage translation strategies, context awareness, and quality diagnosis mechanisms, it improves efficiency while ensuring translation quality, making it a practical tool for handling large-scale subtitle translation tasks.