Changelog

When Cantonese and Traditional Chinese (Taiwan) arrived in Extension 1.4.0 and Desktop 1.2.0, they were translation targets only: SonicMeet could translate into them, but could not caption a meeting held in them.

That gap is now closed. Both are available in the Source language list, so SonicMeet can transcribe a Cantonese or Taiwanese meeting and translate it into any supported target language.

• Cantonese (粵語) — now both a source and a target language
• Traditional Chinese (Taiwan) (繁體中文（台灣）) — the same

As with the translation targets, Traditional Chinese (Taiwan) here means Mandarin as spoken in Taiwan — not Taiwanese Hokkien (台語 / Taigi).

Auto-detect still resolves Chinese to Simplified

Auto detects spoken Chinese as Simplified Chinese. It does not yet tell Cantonese or Taiwan-style Mandarin apart from Mainland Mandarin.

So if your meeting is in either one, choose it explicitly in the Source language list. Leaving Auto on will caption it as Simplified Chinese.

Available in both apps

• Browser Extension 1.4.1 and Desktop 1.2.1
• Both Instant and Accurate translation modes

More on what works today: Cantonese live translation.

SonicMeet now supports two additional translation targets across the browser extension and Windows app:

• Cantonese (粵語): Written Cantonese using Traditional Chinese characters and natural Cantonese wording.
• Traditional Chinese (Taiwan) (繁體中文（台灣）): Taiwan-style written Mandarin using Traditional Chinese characters and local wording. This does not refer to Taiwanese Hokkien (台語 / Taigi).

Available in both apps

• Browser Extension 1.4.0 and Desktop 1.2.0
• Instant and Accurate translation modes

Target languages only

These options are translation outputs, not speech-recognition inputs. Select the language being spoken from the existing Source language list, then choose Cantonese or Traditional Chinese (Taiwan) under Target language.

Full details on what works today: Cantonese live translation.

Version 1.1.3 brings the Accurate translation mode to the Windows app, and fixes a bug that could leave you stuck on the sign-in screen.

Added

• Accurate translation mode. Instant — the ultra-low-latency translation the app has always used — stays the default and is unchanged. Accurate waits for a complete sentence and translates it with a stronger reasoning model, giving it much more of the surrounding conversation to work from. Technical terms land correctly, the phrasing sounds native rather than translated, and pronouns, tense and references stay consistent with what was said a minute ago.
• Where to switch. Open Settings and choose Accurate under Translation mode. You can switch back to Instant at any point, including in the middle of a session.

Which one to use

• Instant — the default. Translation appears with ultra-low latency, which is what you want for live back-and-forth where you have to react immediately.
• Accurate — typically 1–5 seconds behind Instant. Better for technical discussions, presentations, interviews and negotiations, where one mistranslated term costs more than a few seconds of waiting.

Fixed

• Sign-in failure. Fixed an issue that could stop you from signing in and leave you stuck on the login screen. If you have been unable to log in, updating to 1.1.3 resolves it.

Notes

• Instant remains the default — nothing changes unless you switch modes yourself.
• The extra delay depends on sentence length: short exchanges stay near the low end of the range, long uninterrupted passages near the high end.

Version 1.3.0 adds a second translation mode. Until now every translation went through Instant mode, which is built for the lowest latency possible. Instant stays the default for everyone and nothing about it changes. Accurate mode is there for the meetings where getting the wording right matters more than seeing it the moment someone speaks.

Added

• Accurate translation mode. Instead of translating fragments as they arrive, Accurate mode waits for a complete sentence and translates it with a stronger reasoning model, giving it much more of the surrounding conversation to work from. The result reads the way a person would have written it: the right technical term, phrasing that sounds native rather than translated, and pronouns, tense and references that stay consistent with what was said a minute ago.
• Switch it from the captions. Open Settings in the caption area and choose Accurate under Translation mode. You can switch back to Instant at any point, including in the middle of a meeting.

Which one to use

• Instant — the default. Translation appears with ultra-low latency, which is what you want for live back-and-forth where you have to react immediately.
• Accurate — typically 1–5 seconds behind Instant. Better for technical discussions, presentations, interviews and negotiations, where one mistranslated term costs more than a few seconds of waiting.

Notes

• Instant remains the default for all users — nothing changes unless you switch modes yourself.
• The extra delay depends on sentence length: short exchanges stay near the low end of the range, long uninterrupted passages near the high end.

Version 1.2.4 is a focused stability update for Transcribe mode.

Fixed

• No more occasional duplicate original text. Fixed an issue that could cause the original transcript to appear twice in Transcribe mode. Captions now update once as expected.

Version 1.2.3 focuses on the part of SonicMeet you look at the most: the captions themselves. The recognition model behind them has been upgraded, audio now reaches it faster, and you can choose a new way to read the original and the translation side by side.

Improved

• More accurate recognition. An upgraded speech recognition model improves word accuracy, sentence segmentation, and punctuation, especially in noisy rooms, on accented speech, and around technical terms and product names.
• Faster captions. A more efficient audio transmission pipeline sends less data per second of speech and reduces the delay between someone speaking and the caption appearing, without lowering recognition quality.
• Steadier under weak networks. Because each stream carries less data, captions hold up better on unstable Wi-Fi and degrade more gracefully instead of dropping out.

Added

• Bilingual layout: By language. In addition to the existing sentence-by-sentence view, you can now group captions by language: the original transcript stays in one pane and the translation in the other, each scrolling as a continuous block of text. Switch between By sentence and By language in Caption Settings → Bilingual Layout.

• Reading a full language at a time. By language is designed for the moments when you want to read one language continuously, follow the original speech, or copy a clean single-language passage without the alternating lines getting in the way.

Notes

• Bilingual layout only applies when captions are in Bilingual mode; Original and Translation views are unchanged.
• Your bilingual layout choice, along with the rest of your caption display settings, is remembered between meetings.

Version 1.2.2 is one of our biggest updates yet. It introduces a lighter way to follow captions, interactive meeting Q&A with Ask AI, and a more complete experience for reviewing, exporting, and sharing meetings.

Added

• Compact caption bar. Collapse the full panel into a streamlined bar that keeps current captions visible while covering much less of your meeting. Expand it again whenever you need the full transcript, notes, or AI tools.

• Ask AI for meeting Q&A. Ask about key points, decisions, action items, or unresolved questions while a meeting is in progress. After the meeting, open the saved session to keep asking questions and follow up on earlier answers.
• Answers with transcript references. Ask AI grounds its answers in the meeting transcript and includes references you can click to jump directly to the relevant passage.

• Post-meeting recap card. After you finish a meeting, quickly review its duration and content, generate an AI Summary, open the full session, or move straight to sharing and exporting.
• Shareable meeting links. Share the AI Summary, the transcript, or both. Manage all shared links from your history and stop sharing at any time.
• Flexible meeting exports. Export meetings as Markdown or TXT, with options to include timestamps, notes, and the AI Summary.

Improved

• More reliable long meetings. Pause, resume, finish, and reconnection flows have been strengthened to reduce stuck controls, stalled connections, and incorrect session states during longer meetings.
• More accurate meeting duration. Time spent paused, reconnecting, or waiting for a connection is no longer counted as active meeting time.
• More natural caption formatting. Improved English spacing, punctuation, and mixed-language text joining make live captions and exported transcripts easier to read.
• More reliable downloads. Transcript, summary, and recap downloads now use a more dependable flow and show a clear message if a download fails.
• Cleaner history. Empty sessions without usable captions no longer clutter your meeting history.
• Saved display preferences. SonicMeet now remembers the caption window size and display settings, with additional larger font-size options.

Notes

• Ask AI answers questions using the meeting transcript captured so far.
• Ask AI and AI Summary share the same monthly AI Credits.
• Anyone with a public share link can view the meeting content included in that link.

This update polishes the caption reading experience and gives you direct control over which microphone SonicMeet listens to.

Improved

• Clearer caption timestamps. Caption timestamps have been refined so timing information is easier to read while following a live meeting.
• Better caption display. Caption rendering has been optimized for a cleaner, more comfortable viewing experience.

Added

• Microphone input selection. You can now choose the microphone input source in caption settings. SonicMeet still uses the system default microphone by default, but you can manually switch to another input at any time.

This update adds two new ways to control what your captions show and how accurately they capture the words that matter to you.

Added

• Multilingual profanity filter. Mask profane words in captions across multiple languages. When the filter is on, choose between two modes: filter live captions only, or filter both live captions and saved transcripts.
• Custom vocabulary. Add your own words — product names, people, brand names, company names, or domain-specific terms — so SonicMeet recognizes them more accurately. On Team and Enterprise plans, admins and owners can manage a shared vocabulary list at the organization level, which automatically applies to every member.

Notes

• The profanity filter is off by default.

This update improves the caption experience in tighter layouts, especially for RTL languages.

Improved

• RTL display has been refined for a better reading experience.
• The caption area now enters an ultra-compact mode when resized to its smallest size.
• Captions can occupy less screen width while staying comfortable to read.

Improved team management for Team and Enterprise workspaces.

Overlay interactions have been improved for a smoother experience when dragging and resizing.

Improved

• Drag and resize interactions on the overlay are now more responsive and reliable.
• Compact overlay behavior has been refined for better usability.
• Overall overlay responsiveness has been improved across different screen sizes.

Live captions can now include your own microphone input. Before this update, SonicMeet only captioned audio from the active tab.

Added

• Your voice can now appear in live captions.
• SonicMeet can capture both tab audio and microphone audio in the same session.

Notes

• Tab audio captioning works as before.
• Microphone permission is required.