Skip to main content
PromptQuorumBuilt for humans. Structured for AI.
Home/Power Local LLM
Voice, Speech & Multimodal

Subtitle Edit Review: Local Whisper Transcription and Auto-Translate

·7 min readBy Hans Kuepper · Founder of PromptQuorum · Discovery engine for open-weight & open-source AI

Subtitle Edit (stable version 5.3.0, released on 14 October 2026) is a free, MIT-licensed subtitle editor for Windows, macOS and Linux that can run Whisper-family speech-to-text and local-LLM auto-translate on your own computer. The AI engines are optional downloads, and the cloud engines it also lists send data to the provider you pick. Every statement here comes from the project's own pages, not from testing.

Key Takeaways

  • Maintenance: a pre-release, version 5.4.0-beta6, was published on 24 October 2026 on top of the stable 5.3.0, so no stale-project warning applies.
  • Speech to text: Whisper-family and other engines that run on the machine, plus a few online ones.
  • Translation: local llama.cpp, Ollama or LM Studio models, or cloud APIs with your own key.
  • Cost: free under the MIT license, with no ads or limits according to the project's FAQ.
  • Maker: Nikse with contributors, with the full source code on GitHub.

📍 In One Sentence

Subtitle Edit is an open-source desktop subtitle editor whose speech-to-text and auto-translate dialogs can use engines that run on your own computer.

💬 In Plain Terms

You open a video, pick a Whisper-style engine and a model, get a timed subtitle file, and then translate it with a local model or an online service of your choice.

📌Note: This review is based on the GitHub repository, its release notes and the project documentation, checked on 24 October 2026. PromptQuorum has not installed or run the app, so transcription and translation quality are not assessed.

Read in:
🇺🇸en🇩🇪de🇫🇷fr🇯🇵ja🇨🇳zh🇪🇸es🇧🇷pt🇸🇦ar🇰🇷ko

Subtitle Edit is a free, open-source subtitle editor for Windows, macOS and Linux from Nikse and a group of contributors. Besides editing and converting subtitle files, it can transcribe a video with Whisper-family engines and translate subtitles with local or cloud models, which is the part this review covers. It is based on the project's GitHub page, release notes and documentation, checked on 24 October 2026; PromptQuorum has not tested the app hands-on.

Subtitle Edit Review: Local Whisper Transcription and Auto-Translate

What Is Subtitle Edit?

Subtitle Edit is a desktop editor for creating, synchronizing, converting and translating subtitles, and its newer versions add AI features around that core. The project FAQ says it supports more than 380 subtitle formats, and the Windows, macOS and Linux builds are published on the project's GitHub Releases page.

The AI parts are separate from the editor: speech to text, auto-translate, optical character recognition (OCR) and text to speech each use engines that the app downloads on first use or connects to over an API. This review concentrates on speech to text and auto-translate; the documentation calls the current generation Subtitle Edit 5.

Get It

Subtitle Edit is a free download for Windows, macOS and Linux, with an installer or disk image on every platform.

Option
How to get it
WindowsInstaller (Setup.exe) or portable zip for x64; zip for ARM64, from GitHub Releases
macOS.dmg for Apple silicon or Intel, drag to Applications; signed and notarized per the README
LinuxFlatpak (bundles mpv and ffmpeg) or tar.gz for x64 and ARM64
Websitenikse.dk/subtitleedit
Privacy policyNo separate page; the Privacy section of the README is the maker's statement
Source codegithub.com/SubtitleEdit/subtitleedit, MIT license

Version reviewed: 5.3.0, the release GitHub marks as the latest stable one, published on 14 October 2026; beta builds of 5.4.0 are marked as pre-releases. The 5.3.0 files are 59 to 155 MB before any engine or model download.

Requirements

The README asks for Windows 10 version 22H2 or newer, macOS 12 or newer (macOS 14 recommended) and, on Linux, mpv and ffmpeg for the tar.gz builds.

Operating system

What the sources say:
Windows 10 22H2 or newer; macOS 12 or newer, 14 or newer recommended; Linux per the README package list

Linux packages

What the sources say:
The tar.gz builds need mpv and ffmpeg from the package manager; the Flatpak and the macOS .dmg bundle them

Engine downloads

What the sources say:
Engines and models are fetched on first use; the Crisp ASR model list runs from 75 MB to 9.4 GB per model

Translation models

What the sources say:
The curated llama.cpp models are mostly under 8 GB and the largest is about 12 GB

GPU acceleration

What the sources say:
Whisper CPP offers CPU, NVIDIA cuBLAS and Vulkan on Windows; the changelog names Metal on macOS and Vulkan on Linux

Internet

What the sources say:
Needed for engine and model downloads and for any online engine; local engines then run offline

The documentation gives no RAM or VRAM minimum for the app; needs depend on the engine and model you choose, and a "CUDA out of memory" error is answered with a smaller model.

How to Get Started

The steps follow the project's documentation; PromptQuorum has not run them.

  1. 1
    Install and open a video
    Why it matters: Run the installer, dmg or Flatpak, then open a video file or an existing subtitle in Subtitle Edit.
  2. 2
    Transcribe with Video, Speech to text
    Why it matters: Choose an engine and a model, set the audio language or auto-detect, and let the app download the engine on first use before you click Transcribe.
  3. 3
    Check the result
    Why it matters: Adjust post-processing, read the quality report that flags too-short, too-long, overlapping or repeated lines, and fix the grid by hand.
  4. 4
    Translate with Translate, Auto-translate
    Why it matters: Pick an engine and the language pair, translate, review the original and translated text side by side, and confirm with OK.

Features Confirmed by the Sources

Every item below comes from the project documentation or release notes; none has been tested by PromptQuorum.

  • Speech-to-text engines. Whisper CPP, Purfview Faster Whisper XXL, Whisper CTranslate2, Whisper Const-me (Windows), WhisperX and the Python Whisper workflow, plus Qwen3 ASR CPP and Crisp ASR with backends such as Parakeet and Canary; online options are OpenRouter, Alibaba Qwen3-ASR and Google Cloud Speech-to-Text.
  • Transcription helpers. Batch mode, automatic language detection, voice activity detection settings, optional "Translate to English" for Whisper, post-processing and the quality report.
  • Local translation. A managed llama.cpp server with curated models such as TranslateGemma, Gemma 4 and Qwen, Ollama, LM Studio, and "advanced" variants that translate in batches with context, a synopsis and a glossary; a MADLAD engine covers 419 languages.
  • Cloud translation. Google Translate, Microsoft Bing, DeepL, ChatGPT, Anthropic, Gemini and others, mostly with your own API key, and an OpenAI-compatible engine for any server that speaks that API.
  • Around the core. A video player with waveform, OCR for image subtitles, text to speech and generated background music, as listed in the documentation.

Data Flow and Privacy

The README says Subtitle Edit does not collect, store, transmit or analyze your subtitle files, media files or metadata, and that editing, conversion, playback and local auto-backup run on your device. Local engines such as Whisper CPP, llama.cpp, Ollama and LM Studio process audio or text on your machine, or on a server you point the app at.

The exception is what you switch on yourself. Per the README, optional online services for translation, speech to text, text to speech, OCR or dictionary lookups receive only the data needed for that request, directly from the app, under the provider's own privacy policy. The engine descriptions mark Google Translate, DeepL, ChatGPT, Anthropic, Gemini and the online speech engines as services that need internet access, most of them an API key. The OpenAI-compatible and llama.cpp "remote server" settings send text to whatever address you enter.

📌Note: These are the maker's declarations, not audit results. The code is open, but PromptQuorum has not read it or inspected the app's network traffic, and the hosts the engines and models are downloaded from were not checked.

Who Should Use It

  • Subtitlers and video editors who want transcription, timing and translation in one editor instead of several tools.
  • Privacy-minded creators who prefer a local Whisper engine and a local translation model over uploading media.
  • Anyone handling a whole series of episodes, who needs the same conversion or translation applied to every file.

What We Could Not Verify

  • Accuracy and quality. PromptQuorum did not run any speech or translation engine, so word error rates, hallucinations and translation quality are not assessed.
  • Speed and hardware. No timings or memory figures were measured; the effect of the CUDA, Vulkan and Metal backends on real files is untested.
  • The official website text. The page at nikse.dk is rendered with JavaScript and was not read; facts here come from GitHub and the documentation site.
  • Download sources. The documentation says engines and models download automatically, but which servers they come from was not checked.
  • Pre-release stability. Version 5.4.0 is still in beta, and the 5.3.0 stable line was itself preceded by dozens of beta and release-candidate builds, so behavior may change quickly.

Competitors and Alternatives

faster-whisper

Platforms:
Windows, macOS, Linux
Price / license:
Free / MIT
Key difference:
Python library for fast Whisper transcription; you build the subtitle workflow yourself
Articles about faster-whisper (5)

Also mentioned in:

+6 more not shown

MacWhisper

Platforms:
macOS
Price / license:
Freemium / proprietary
Key difference:
Mac app focused on transcription with a one-time purchase; Subtitle Edit adds format editing and engine choice
Articles about MacWhisper (2)

Also mentioned in:

TranscriptionSuite

Platforms:
Windows, macOS, Linux
Price / license:
Free / GPL-3.0
Key difference:
Local transcription with live mode, speaker labels and an audio notebook, per its directory entry
Articles about TranscriptionSuite (1)

Confirm each app's current price, license and platforms on its own page before choosing.

Compare in directory

Frequently Asked Questions

Do I need internet for the first transcription?

Yes, once: the app downloads the chosen engine and model on first use. The third-party components page also describes placing the files by hand in the app's data folder, which on Windows is %APPDATA%\Subtitle Edit and on Linux and macOS ~/.config/Subtitle Edit.

Which Whisper engine suits an NVIDIA graphics card?

The documentation recommends the Whisper CPP cuBLAS backend or Purfview Faster Whisper XXL for the fastest transcription on NVIDIA GPUs, and a smaller model if you hit a CUDA out-of-memory error.

Can it label who is speaking?

Yes, with the Crisp ASR engine. "Detect speakers" uses NVIDIA's Sortformer diarization for up to 8 speakers, downloads a 107 MB model on first use and needs Crisp ASR v0.8.37 or newer.

Can I use a translation model that is not in the list?

Yes. Drop a .gguf file into the llama.cpp/models folder or tick "Use remote server" and point to your own llama-server. The docs note that TranslateGemma 27B needs roughly a 16 GB download and about 20 GB of VRAM, and call the 12B model already a strong choice.

Can I run it without the graphical interface?

Yes, through seconv, the command-line converter published with every release. Its auto-translate supports openai-compatible and DeepL engines, a --translate-api-key option and a --translate-prompt option.

Verdict

Subtitle Edit is a long-running, free and openly licensed editor that now bundles local Whisper-family transcription and local-LLM translation next to its classic subtitle tools, and its change log shows four pre-releases between 20 and 24 October 2026. Against that, the AI engines are separate downloads of up to several gigabytes, cloud engines send your text or audio to third parties, the newest line is still in beta, and nothing about output quality could be checked here. It suits people who already work with subtitles and want one place to transcribe and translate on their own machine; anyone who only needs a transcript may find a dedicated transcription tool simpler.

Sources

← Back to Power Local LLM

MCP