VideoCaptioner

The interesting engineering here is not the speech model — it is the layer after it. Raw ASR output arrives as long, badly broken sentences; VideoCaptioner routes that text through an LLM configured by you, which re-segments it semantically and cleans the errors. In practice that is the difference between subtitles people tolerate and subtitles people read without noticing. Three configuration choices drive almost all of your quality. First, the ASR engine: FasterWhisper locally supports all 99 languages and is the vendor's own top recommendation, with the guidance being Medium or larger for Chinese, Small for English, and Large-v2 for other languages. Second, the LLM: the app bundles a basic model, but pointing it at gpt-4o or gemini-2.0-flash-exp measurably improves segmentation and correction, and Ollama with llama3.1:8b gives you a fully local, zero-cost pipeline. Third, the script hint — paste your terminology list or source script and the correction pass gets noticeably sharper on names and jargon. The honesty section matters too: the vendor's own troubleshooting notes say hallucination and repetition are solved with VAD filtering, a larger model, Large-v2 instead of Large-v3, or audio separation in noisy rooms, and they advise checking your API key, base URL and thread count when LLM calls fail. Those are real operational details no marketing page prints. The weaknesses are structural, not incidental. It is desktop software: Windows 10/11, macOS 10.15+, or Ubuntu 20.04+ / Debian 11+ / Fedora 35+, with Python 3.10+ if you run from source and 8GB+ RAM if you use local Whisper. A Docker build exists but the vendor explicitly calls it basic and experimental. There is no timeline editor and no team collaboration layer, so it does not replace a post-production suite. Batch processing covers volume, but it is queue-based, not an automation API. And because your LLM quality is your API key's quality, the cost and reliability of the semantic layer are yours to manage. For a solo creator, a lecturer, or a small studio captioning in bulk, that trade is a good one: free software, local privacy, and captions that look hand-polished.