Skip to content

Video & meeting translation ​

Read original captions and translations in YouTube, X, Udemy, Disney+, and Google Meet, Teams, or Zoom web meetings. Most platforms need readable captions; X also offers local AI transcription.

VIDEO · BILINGUAL CAPTIONSWalkthrough
Video / meeting caption example

我们一起回顾一下计划。

Auto demo · sample content

A video with subtitles ​

  1. Enable Video subtitle translation in the extension menu or settings, and turn on the video's native captions.
  2. In YouTube or X, click the FluentRead icon near the fullscreen button.
  3. Choose Bilingual / Translated / Original. Other supported platforms provide display controls near their captions.

The service follows webpage translation by default; choose a separate one in Video translation settings. Selecting Off in the player disables video translation and removes its button. Re-enable it from the extension menu or settings.

Translation depends on the selected service's response. Opening a video or seeking to an untranslated section can cause a delay. Retry a failed translation without repeating transcription: recognized source captions and their timeline remain available.

Meeting captions and human subtitles ​

Meetings use web clients, including Zoom's /wc/ client. The platform, host, or account must allow captions. Browser extensions do not control desktop meeting apps.

Two options under Video translation settings → General are on by default:

  • Automatically enable bilingual meeting captions attempts to enable captions supplied by Google Meet, Teams, and Zoom. Enable native captions manually if it fails. Disabling translation or hiding FluentRead captions restores the native display.
  • Prefer human subtitles uses readable target-language human captions on YouTube, Udemy, and Disney+. Missing or inaccessible portions fall back to translating source captions.

FluentRead does not start meeting recording or transcription, or generate captions for meetings that provide none. Udemy and Disney+ availability also depends on account, region, and player restrictions.

Display and download ​

Adjust size, background, position, width, and bilingual order in video appearance settings. Its example only previews styling; it reads no video and sends no requests.

If captions are early or late, open Subtitle timing in the player menu (under Subtitle options on X). Adjust by 0.5 seconds, up to ±10 seconds, or reset to zero. The saved offset applies to later videos and moves both lines together. It does not change playback or exported timestamps. Timing needs an available subtitle timeline.

Click Download subtitles in the menu header, then choose Original / Translated / Bilingual to save SRT. Each bilingual cue puts the original on the first line and the translation on the second. Identical lines appear once. The bilingual export uses a single timeline. An overlapping cue ends when the next one starts, and cues that start together share one entry.

Translated or bilingual exports first show how much text still needs translation and its time span:

  • Translate missing text and export complete subtitles requests missing translations only after confirmation. Usage follows your selected service.
  • Export available results only skips untranslated cues, adds partial to the filename, and reports how many were skipped.

Close the menu or press Esc to cancel while translating missing text; completed translations remain reusable. Changing the video or translation settings requires a new preview. X AI subtitles become exportable only after complete recognition succeeds.

An X video without subtitles

An X video without subtitles ​

The desktop Chrome, Edge, and Firefox extensions use Whisper in the browser by default to transcribe X videos.

  1. Open the X player's FluentRead menu and choose to generate AI subtitles.
  2. For Whisper in the browser, select and confirm a model download on first use. Small is recommended (about 590 MB). Tiny (about 100 MB) and Base (about 150 MB) use less storage.
  3. Wait while the extension reads and transcribes the audio. You can stop at any time. Stable sentences appear early, and the timeline is cached and becomes exportable after recognition succeeds.

All Whisper models support multiple languages. Use Auto when the language is unknown or changes, or specify a known single language. Whisper in the browser uses available GPU acceleration and falls back to CPU. FluentRead sends recognized text to your translation service according to the display mode. Original-only mode displays the transcript locally.

The limit is 20 minutes per video. Speed depends on the video and device. Some media formats or restricted videos cannot be read; try opening the post or refreshing. Quiet speech, music, accents, and specialist terms can affect accuracy.

Use Subtitle options → Recognition model / Regenerate this video to change models or recognize again. Existing captions remain until confirmation; cancel preserves them. Stop an active download or recognition before switching models. Regeneration does not require clearing other videos' caches.

Manage Whisper models under Video translation settings → Local AI subtitles. The first download requires a connection. Retry failed downloads to reuse complete files and fill missing ones. Clearing browser caches can require downloading again. A recognition failure removes incomplete previews, while a translation failure preserves the recognized source.

Browsers that support the File System Access API can also enable Share models across browsers. Select and authorize the same folder in each browser; models are stored in its FluentRead/whisper subfolder. Existing models can be moved there. Disconnecting does not delete shared files. Browser-local extension caches remain available.

Use Local Qwen3 ASR ​

If you have vLLM with support for this model, run the command below to start a local service. It must serve the fixed model Qwen/Qwen3-ASR-1.7B at POST /v1/audio/transcriptions without credentials. vLLM model recipe

sh
vllm serve Qwen/Qwen3-ASR-1.7B --host 127.0.0.1 --port 8000

In Video translation settings → Local AI subtitles → Recognition method, choose Local Qwen3 ASR, enter http://127.0.0.1:8000, and save. The address must use 127.0.0.1 or [::1]. Once the service is running, select Generate AI subtitles on an X video with a readable audio track. This sends the audio to your local service.

FluentRead estimates caption timing from audio windows. Stopping cancels the extension's request, and the server may continue processing audio it already received. If the service fails, retry or switch back to Whisper in the browser in settings.

Reading annotations and word highlighting

Reading annotations ​

In the browser extension, enable Video appearance → Reading annotations or its header icon in the YouTube/X menu. It is off by default and unavailable in the userscript. Supported readings include Japanese romaji, Korean romanization, Chinese pinyin, and Latin transliteration for Russian, Greek, Arabic, Hindi, and Thai.

Readings follow the original text and stay hidden in translation-only or hidden-caption modes. Enabling them does not change display mode. Japanese first needs about 17.8 MB of dictionary data, Hindi about 0.5 MB. Verified data is cached locally for offline reuse; retry failed downloads in the menu. Captions are not sent to a reading service, and originals and translations can display while data is prepared.

Readings are aids, and names, alternate pronunciations, and sound changes can be wrong. Check the audio. Annotations do not change source captions or exports.

Word-by-word highlight ​

Enable Word-by-word highlight in video appearance settings (off by default) to light up YouTube and X captions as they play. Timing is estimated from the whole cue's duration and text length, so it may differ from actual pronunciation.

Sentence learning

Sentence learning ​

Choose Start sentence learning in a YouTube/X subtitle menu to read the source sentence, context, and whole-sentence translation. It needs a readable timeline; X AI subtitles must finish first. Long or unpunctuated captions may be labelled as fragments, rather than complete sentences.

The learning panel splits existing captions into sentences using local rules. Each sentence keeps its source text and original caption timing. Sentences from the same caption can share a playback range, so moving between them keeps your current position.

  • Previous / Next select the adjacent sentence. Moving to a different source passage seeks to its start and preserves the current play or pause state.
  • Replay source passage plays from the start of the passage. Loop source passage / Pause at passage end follow the original caption timing. If the playhead is outside the range, use Replay to start. Manual seeking ends playback assistance and preserves your playback speed.
  • Explain is optional and calls your selection AI explanation service only when requested. It can use neighboring captions if context is allowed.
  • Edit changes the translation or explanation and pins the sentence. Drafts survive sentence changes during this panel session; return to the current sentence to follow playback again.
  • Save sentence keeps the source, confirmed translation, explanation, context, and time in the Learning Center. No explanation is required. Wait for a pending translation or enter your own before saving.

With focus in the player, arrows change sentence, R toggles looping, and Q toggles pause at passage end. Text fields keep normal typing; an open subtitle menu takes keyboard priority. Drag the panel title to move it or use its style button to change layout and size. Closing learning or captions, or switching videos, stops playback assistance.

Learning Center sources reopen the video near its original time; X positioning needs the extension enabled. Saved text remains if a video is deleted or restricted. Subtitle offsets and bilingual order do not change the original times used for saving and looping. Vocabulary exports that omit private context also omit video URLs and times.

Missing or inaccurate subtitles ​

Enable platform captions and wait for them to load. Ordinary translation cannot start without captions; on X, try local AI generation instead.

An empty transcription means recognition found no readable speech. Check the audio and source language, or try Small instead of Tiny/Base. For a translation failure, retry or change translation services. Check important information against source captions and audio.

Closer languages. A bigger world. · Privacy policy