From transcription
to Japanese subtitles.
A Mac app for making Japanese subtitles for videos in foreign languages. Open a video, and KotobaSync transcribes the speech with Whisper on your Mac, translates it into Japanese with the translation engine you choose, and lets you check and fix it while the video plays before saving it as SRT. You can also open any video and edit its subtitles.
The buttons, menus and settings are in Japanese, and translations are into Japanese.
- Version 1.2.75
- 14.3 MB
- Notarized by Apple
- macOS 15 or later
- Apple silicon (M1 or later)
次にすること:提案を確かめて、日本語字幕を保存します
もちろん行くよ
字幕を直す
I wouldn't miss it.
任意の確認:● 提案 1件
00:01:12.400
Are you coming tonight?
今夜 来るの?
00:01:14.900再生中
I wouldn't miss it.
もちろん行くよ
00:01:17.300● 提案
Meet me at the station at seven.
駅で7時に会いましょう
00:01:20.150
Don't be late this time.
今度は遅れないで
Open a video, then press the buttons from 1 to 4 in order.

What to check before you download
Seven points: whether it runs on your Mac, what it connects to, and where data is kept.
It needs macOS 15 or later and an Apple silicon Mac.

- Cost
- FreeNo account needed. If you choose Codex for translation, it uses your own ChatGPT / Codex usage allowance
- Requirements
- macOS 15 or later, Apple silicon (M1 or later)Not for macOS 14 or earlier, or Intel Macs. A translation engine is not included (see below)
- App interface
- JapaneseThe spoken language is detected automatically (you can also set it). Translations are into Japanese
- Permissions
- None by defaultSpeech Recognition is asked for only if you choose Apple Speech for transcription. It reads and writes only the video and subtitle files you choose in Open and Save dialogs
- Network
- Four kinds of connectionModel download from Hugging Face (the transcription model, and the speaker analysis model only if you prepare it); OpenAI only if you choose Codex for translation (Ollama and LM Studio stay on this Mac; still images from the video are sent only if you turn on the optional Codex image analysis); looking up the work on Wikipedia only if you say yes (the title and year only), and on TMDb only if you enter your own TMDb API key; a daily update check with k386.sub.jp (can be turned off). Details
- Where data is kept
- Only on this MacSaved SRT files go next to the video by default. Models, autosaves and translation progress are kept in
~/Library/Application Support/SubtitleStudio - Latest version
- 1.2.75 (November 10, 2026)When reading subtitles burned into the picture, other on-screen text (signs, number plates, logos, titles drawn letter by letter, end credits) is told apart by its size, position, colour and movement and removed, with one click to bring it back. Reading now runs on the CPU from the first frame, which stops it from hanging.
Features
There are three buttons at the top of the window: 1 動画を開く (Open Video) → 2 日本語字幕を作る (Make Japanese subtitles) → 3 確認して保存 (Check and save). One press of 2 transcribes the speech if there are no subtitles yet, builds the work notes and glossary, translates everything and runs rule checks on names and line length; its menu has transcribe-only, a 5-line trial translation and the work notes. 3 shows the rows worth checking before you save (確認して保存(要確認 N行)), or saves the Japanese SRT when there are none; its menu also has AI rereading, the detailed checks, Watch through and video export. A line under the buttons tells you what to do next.
Transcribe on your Mac
Whisper (whisper.cpp 1.8.2, included) runs on your Mac. The spoken language is detected automatically. Silero VAD (included) finds where people are speaking and moves subtitle times closer to the voice, while leaving enough time to read them. Apple Speech is also available. Long videos are processed in parts, with progress and a stop button. When imported subtitles leave out some speech (English subtitles of a Korean drama often skip the scenes spoken in English), the app can transcribe just those stretches and add them as lines to check.
Translate with the engine you choose
Choose Ollama or LM Studio (on this Mac) or Codex App Server (OpenAI's cloud). It translates the whole video without any character setup first. Ollama and LM Studio translate 8 lines at a time with the conversation around them by default (16, 40 or the old one line at a time can be chosen) and then check the meaning and which subtitle each line belongs to. There are four styles. Long one-line translations are split at a natural break, and lines from Codex that run far over their time are asked again in a shorter form. Translated lines are saved one by one, so you can continue from where it stopped.
Review with suggestions
AIに読み直させて提案を作る (Have the AI reread and suggest), in the menu of 3 確認して保存, looks for lines to fix using the lines around them and shows the fixes as suggestions; nothing changes until you accept one. After a translation with Ollama or LM Studio it runs by default; after Codex it does not, since Codex reads 40 lines at a time and its rereads were more often worse. It uses the macOS dictionary (offline) and a translation back into English to check the meaning. A glossary keeps names consistent across the whole video. The lines still to look at are counted above the list in two rows: 保存前に (Before saving — rows that would otherwise appear in English or mistranslated) and 任意の確認 (Optional — import mismatches, suggestions). Click a count to see only those rows, what to do with them and how many are done; the 確認 (Review) menu lets you work through them with ⌘↩ (confirm/accept and next), ⇧⌘⌫ (reject/undo and next), ⌘] / ⌘[ (next/previous open row), R (play this row) and ⇧⌘R (play the open rows one after another).
Edit while the video plays
Uses the same keys as YouTube (Space/K, J/L, ←/→ and more). The list follows the subtitle being played, and the source and translation are shown together. You can shift all subtitles from a selected one to the current position, and undo up to 20 times. Before saving or exporting, it tells you if any rows have no translation and would appear in English. Translation style is remembered per video.
Use subtitles already in the video
When you open an MKV file, its built-in subtitles (source and Japanese) are loaded. Image subtitles (PGS, VobSub) and subtitles burned into the picture can be read as text (OCR). MKV, WebM and similar files are converted to MP4 with ffmpeg on your Mac; the original file is not changed.
Save as SRT, or export a video
Saves SRT (UTF-8) for the source and for the translation. It can also export an M4V video with the subtitles burned in or added as subtitle tracks. If Movie Metadata is installed, the exported video can be opened in it right away to add the work’s details. Unsaved subtitles are saved automatically 3 seconds after a change and can be restored if the app quits unexpectedly.
What it can't do
Making subtitles for films, series, anime, TV programmes or other copyrighted videos and publishing, distributing or selling those subtitles or videos without the rights holder's permission may infringe copyright. Do not use the app that way. KotobaSync is for adding Japanese subtitles to videos such as your own recordings or recorded conferences and talks, to watch them yourself. Commercial use is prohibited. The app asks you to agree to its Terms of Use at first launch; its author accepts no liability for use that breaks them.
Transcription times are estimates, and names or unclear speech need to be fixed by hand. Translation quality depends on the model you use. Play the video and check and edit the translation.
- macOS 14 or earlier and Intel Macs are not supported.
- A translation engine is not included. Install and set up Ollama, LM Studio or Codex yourself. With Ollama or LM Studio you also need to download a model, and your Mac needs enough power (more memory lets you use larger models). For transcription, editing and saving only, no translation engine is needed.
- It does not reliably identify who is speaking or understand the context of the whole work.
- Speed and accuracy on a full two-hour film have not been tested.
- Apple Speech may not be available depending on your Mac and the language.
- Text read from image subtitles and from subtitles burned into the picture may contain errors.
- Converting MKV and similar files, and exporting subtitle tracks, need ffmpeg (
brew install ffmpeg). - Versions before 1.2.0 are not distributed on this site (1.2.0 is the first public version).
Getting started
-
Download and move to Applications
Open the DMG and drag ことばシンク (KotobaSync) to "Applications".
-
Get ready on the welcome screen
The first time you open it, the ことばシンクへようこそ (Welcome) screen shows what is ready (✓) and what still needs setting up (要準備): ① the transcription model, ② the translation engine and ③ ffmpeg (optional). You can download the model from this screen. The standard model, large-v3-turbo, is about 1.6 GB; tiny (about 75 MB) is enough to try it out.
-
Open a video and follow the buttons
Press 1 動画を開く (Open Video), then 2 日本語字幕を作る (Make Japanese subtitles). When 日本語字幕ができました (Japanese subtitles are ready) appears, press 3 確認して保存 (Check and save) to look at the rows it points to and save the SRT, or export a video with the subtitles. To edit existing subtitles, load an SRT file instead of transcribing.
Network and data
KotobaSync connects to the network only in the four cases below. It sends no usage statistics or error reports and shows no ads. Video and audio files are never sent (still images from the video are sent only if you turn on the optional Codex image analysis).
TsumugiWith Ollama or LM Studio, your subtitles stay on this Mac.
| Connection | When | Where to | What is sent | Can you turn it off? |
|---|---|---|---|---|
| 1. Model download | When you press 取得 (Download) for the transcription model in Settings or on the welcome screen, or モデルを準備… (Prepare model) for speaker analysis under 人物設定 (Characters) in the work notes window | Hugging Face (huggingface.co) | Only the request for the model file | It happens only when you press the button. After the download, transcription and speaker analysis need no internet connection |
| 2. Translation engine | When you translate or review translations | Ollama and LM Studio: this Mac only (127.0.0.1). Codex: OpenAI | Ollama and LM Studio: nothing leaves this Mac (proxies, redirects and cloud models are refused). Codex: the subtitle text and the translation instructions, using your ChatGPT / Codex usage allowance. Only if you turn on Codexで映像の画像を補助解析する (Codex image analysis, off by default), also still images from the video (up to 12 sections × 3 frames per run) | Choose Ollama or LM Studio to keep everything on this Mac. Codex asks you to confirm once before the first send; your answer is saved and can be withdrawn in Settings |
| 3. Looking up the work | When a video is opened, only after you answer 調べる (Look it up) to the one-time question | Wikipedia and Wikidata (Wikimedia Foundation) TMDb (The Movie Database), only if you enter your own API key | Only the title and year (and season for a series) from the file name; TMDb also receives the search words, the work’s number and your API key. No subtitle text, audio, video or file location | Yes. Answer 調べない, or turn it off in Settings → 基本 (Basic); remove the TMDb key there to stop using TMDb |
| 4. Update check | Once a day. When a new version exists, it is downloaded and verified in the background and installed when you quit the app (from 1.2.1) | k386.sub.jp (/KotobaSync/api/version.json and the DMG) | Nothing; it only fetches version information and the installer. No subtitles or settings are sent | Yes. Turn off アップデートを自動で確認 (Check for updates automatically) or アップデートを自動でインストール (Install updates automatically) in the app menu |
Where data is kept (only on this Mac)
| Data | Location |
|---|---|
| Saved SRT files, converted MP4 files, exported M4V files | The same folder as the video by default (can be changed in Settings) |
| Transcription models | ~/Library/Application Support/SubtitleStudio/Models |
| Speaker analysis model (only if you prepare it) | ~/Library/Application Support/SubtitleStudio/SpeakerModels |
| Autosaves | ~/Library/Application Support/SubtitleStudio/Autosave (the 20 most recent) |
| Translation progress, glossary, work notes | ~/Library/Application Support/SubtitleStudio/TranslationCheckpoints |
| Converted videos (only when the save folder can't be written to) | ~/Library/Caches/SubtitleStudio/ConvertedVideos |
| Settings | ~/Library/Preferences/com.subtitle.studio.plist |
| LM Studio API token (if you use one) | The macOS keychain (item com.subtitle.studio.lmstudio) |
| TMDb API key (if you use one) | The macOS keychain (item com.subtitle.studio.tmdb) |
- The folder name SubtitleStudio is the app's internal name, kept for compatibility.
- No special permission is needed for normal use. Only video and subtitle files you choose in Open and Save dialogs are read or written.
- The Speech Recognition permission is asked for only if you choose Apple Speech for transcription. Apple Speech is used only when on-device recognition is available for the language; audio is not sent to a server instead.
- When a new version is installed, its SHA-256, Developer ID signature and Apple notarization are checked first. If you have unsaved subtitles, you are asked about them before it relaunches.
FAQ
Why is it free?
It is one of the free Mac apps from K386. There are no ads and no paid version. The app does not include a translation service: Ollama and LM Studio run on your Mac, and Codex uses your own ChatGPT / Codex usage allowance.
Do I need a translation engine?
Only for translating. Transcription, editing and saving SRT work without one. To translate, install and set up one of Ollama, LM Studio (both run on your Mac) or Codex (OpenAI's cloud; sign in to ChatGPT with codex login). On a new Mac the engine is chosen for you (Codex if it answers, otherwise a local server with a model that fits the Mac). You can check or change it in the 基本 (Basic) tab of Settings.
Does it work offline?
Yes, after the transcription model has been downloaded. Transcription, editing and saving run on your Mac, and so does translation with Ollama or LM Studio. Translation with Codex needs an internet connection.
Can I use it just to edit subtitles?
Yes. Open a video and load existing subtitles with 字幕を読み込む (Load Subtitles), or use the subtitles built into an MKV file. You can then edit the text and times while the video plays and save SRT.
What do I need for MKV or WebM files?
ffmpeg on your Mac (brew install ffmpeg). KotobaSync uses it to convert the file to an MP4 that can be played; the original file is not changed. ffmpeg is also needed to export subtitle tracks. Burning subtitles into an M4V video does not need ffmpeg.
Can I use it on an Intel Mac or macOS 14?
No. It needs an Apple silicon (M1 or later) Mac and macOS 15 or later.
How do I update?
From 1.2.1 it updates itself. The app checks k386.sub.jp once a day; when a new version exists it downloads it in the background, checks the file's SHA-256, the Developer ID signature and Apple notarization, and puts it in place the next time you quit the app — it never restarts while you work. To update right away, choose アップデートを確認… (Check for Updates…) in the app menu and then 今すぐ再起動して更新 (Restart and Update Now). Version 1.2.0 shows a notice instead; press 更新して再起動 (Update and Relaunch) there once. New versions are also announced on Updates (also via RSS).
How do I uninstall it?
Quit the app and delete ことばシンク from Applications. If you don't need the data, also delete ~/Library/Application Support/SubtitleStudio (models, autosaves, glossaries), ~/Library/Caches/SubtitleStudio and ~/Library/Preferences/com.subtitle.studio.plist. If you saved an LM Studio API token or a TMDb API key, remove the com.subtitle.studio.lmstudio and com.subtitle.studio.tmdb items in Keychain Access. SRT and video files you saved are not removed.
Read more (in Japanese)
Longer articles about ことばシンク on note and Zenn. All articles
Whisper とローカル LLM で、Mac の中だけで日本語字幕を作るアプリを作った(AI の直しを自動で使わない理由)How KotobaSync keeps transcription and translation on the Mac, and why AI rewrites stay as suggestions
Download KotobaSync
Version 1.2.75, released November 10, 2026. The download page lists the SHA-256 checksum and how to verify the file.
Trademarks
- Apple, Mac, macOS and Apple silicon are trademarks or registered trademarks of Apple Inc. in the United States and other countries and regions.
- Intel is a trademark of Intel Corporation or its subsidiaries.
- OpenAI, ChatGPT, Codex and Whisper are trademarks or registered trademarks of OpenAI.
- LM Studio is a trademark of Element Labs.
- YouTube is a trademark of Google LLC.
- Ollama, Hugging Face and TMDb are trademarks or registered trademarks of their respective owners.
- whisper.cpp (© The ggml authors) and Silero VAD are used under the MIT License. The license texts are included in the app.
- Other company, product and service names on this site are trademarks or registered trademarks of their respective owners. ™ and ® are omitted in the text.
- K386 and KotobaSync are not affiliated with, endorsed by or sponsored by the companies and organizations above.
This product uses the TMDB API but is not endorsed or certified by TMDB. KotobaSync uses it only when you enter your own TMDb API key.