English ↔ Japanese demonstration
Other languages are supported too. Availability depends on the recognition, translation, and speech system. The exact translation may vary.
Web and desktop
- Start live translation, choose your microphone or system audio, and grant access.
- Set speech to English, translation to Japanese, and turn on Bilingual.
- Turn on the speaking-person button beside the bottom timer and test both directions. Purple means on; a slash means off. Each new session starts with autoplay off.
Play one sentence or stop speech
Autoplay reads newly finalized translations in order. It does not replay past history. Hover over a translated row, focus its toolbar with the keyboard, or tap it. Select play beside edit to interrupt automatic speech with that sentence, using the separate fixed speed. This does not enable autoplay. Turning autoplay off also cancels queued speech. Stop a manually selected sentence with its own stop button. Stopping translation also ends audio capture.Let attendees listen on their own devices
The host enables sharing and listener Text-to-speech, then distributes the URL or QR code. Listeners choose a language and select Unmute. See live captions. Volume is 0–100% (default 100%). Speed defaults to Auto; explicitly choose Manual for 0.5–2.0× playback. Volume and speed settings are saved on that device. Mute is a separate control. If the browser cannot select an output, use device settings. After an app interruption, return to the page and unmute if needed. If you also hear the bot, stop one output to avoid duplicate speech.Meeting bot
- Open Meeting Logs → Web Meeting Bot → Schedule and enter the meeting URL and time.
- Choose English as the first language and Japanese as the second. Enable Interpretation mode, check volume, and schedule. Web initially uses off and 70%; later bookings restore your last submitted settings.
- Have the host admit and unmute the bot. Confirm that speech is ready, then test both directions.
Live controls, chat, and calendar settings
The creator can use the interpretation switch in Live, or edit Setting → Save. Mobile also offers Interpretation TTS mode / Audio volume. A newly called mobile bot starts with interpretation off and volume 100%.
Turning interpretation off stops active and queued speech while captions continue. Re-enabling starts with new speech. Volume cannot be changed while off, but its value is retained.
Acknowledgment is separate from speech readiness. Resolve any reported failure and resend
tts on; also check whether the host muted the bot.
Integration keeps separate interpretation, language, and volume defaults for Google and Outlook Calendar. They apply to upcoming meetings; initial volume is 70%. The second language remains available for captions when interpretation is off. See Web Meeting Bot for scheduling, permissions, and usage limits.
Automatic detection, supported languages, and captions
Both fixed languages must support bot speech. Check the second language if it was filled automatically. This pair determines the spoken translation direction. With first language = Automatic, second = Japanese, the other speaker is translated into Japanese. Japanese replies go to the most recently confirmed counterpart language. Let the other person speak first. A language assumed by fallback recognition is not remembered as the reply destination. Unsupported speech languages can still appear as captions. Send each command as a whole chat message:en ja sets a fixed pair, auto en ja sets detection candidates (up to four), and auto restores the creator’s defaults. Default candidates follow personal or workspace settings.
The bot camera can show original text, translations, and a listener QR code. Captions continue with TTS off. Listener speech has separate controls.
Offline mobile
- While online, open Home → Offline → Offline Models. Prepare recognition and translation models plus voices for both languages. Check remaining plan time and expiry.
- Select Start, allow the microphone, and choose English → Japanese with Bilingual on. Bilingual defaults to off and is saved.
- Enable autoplay to the right of the bottom microphone and test both directions. It defaults to off but is saved, so check its current state each time.
Languages, downloads, and usage time
Recognition models v0.1/v0.2 support English, Japanese, Korean, Chinese (Simplified/Traditional), Cantonese, Vietnamese, and Thai. Version v0.0 excludes Vietnamese and Thai. Translation targets are the seven languages/script options excluding Cantonese. Choose an explicit offline pair; online Automatic detection is unavailable here. Simplified and Traditional Chinese are script variants, not separate spoken languages for a bilingual pair. Your translation provider may require models for both source and target. Optional VoicePing translation models are not installed automatically. After installing device voices, return to the app and confirm readiness. Sync your entitlement while connected, then test without a connection. Free use has a daily time limit; unlimited use requires a valid eligible paid plan.Translate another app’s audio and save history
From Home → Offline, select Screen Record on iOS or Device Audio on Android 10+. The session opens stopped. Press record, grant OS consent, then play the source app. Enable autoplay to hear translations. This processes audio and does not save screen video. This iOS route does not use the microphone. Android needs recording permission and capture consent. Some apps restrict capture; streaming an online video still needs a connection. End the session and re-enter to change input source. Restarting capture requires OS consent again. Back stops capture before opening the title/save screen; canceling save leaves it stopped. If stopping fails, retry and confirm the OS recording indicator disappears. History and recordings are stored on the device. Check saved content before deleting data or uninstalling. Background buffering is finite; processing cannot continue indefinitely after iOS suspends the app. See offline translation.Speech and automatic audio adjustment
Computer, listener, bot, and mobile controls are independent. For silence, check the intended output’s switch, volume, supported voice, and device. Pause between short phrases if speech falls behind. For repeated translations, try headphones or lower volume.
Automatic speed and manual settings
Actual mobile speed varies with iOS, Android, and the chosen voice. Manual fixed speed remains separate: automatic speed does not overwrite it or multiply the two rates.
Recognition, translation, and speech preparation take time. Speech queues have limits, and simultaneous speakers are not guaranteed to be recognized accurately.
Bot volume correction and mobile output
Bot volume ranges from 0–100% and applies to active and queued speech. A curve gives finer control at low volume; an additional 10 dB output reduction applies even to saved settings and 100%. Automatic gain control (AGC) is disabled to prevent the level from rising again. This does not normalize loudness for every participant. Mobile device-audio playback does not deliberately lower the source app’s volume or switch it to a call mode. iOS mixes speech with the source. Check the available output route and device volume.Echo protection and stopping speech
- Web: Uses supported microphone echo cancellation and exclusion of its own output. A fallback may filter text similar to recent TTS; it can also filter a person repeating that phrase.
- Bot: Uses the actual output audio to remove feedback, then suppresses residual repeats. It cannot guarantee removal of all echoes from participant devices.
- Mobile app: If required protection or an audio route becomes unavailable, speech is canceled while healthy capture continues. Check the notice and output before retrying. Manual playback with autoplay off may enter a nearby microphone. Device capture uses dedicated output exclusion.
Recognition accuracy, dictionaries, and noise adjustment
- High Accuracy: Mobile recognition quality is independent of autoplay. Offline autoplay also leaves online recognition preferences unchanged.
- Bot recognition segments: Maximum message duration defaults to 10 seconds, configurable from 5–20. Interpretation does not force it to two seconds.
- Recognition recovery: A fallback service may recognize a specific language while Automatic remains selected.
- Dictionaries: Recognition hints and translation terms update during meetings. The pronunciation dictionary does not control TTS pronunciation.
- Quiet speech: Mobile bounds ambient-level calibration to reduce missed quiet speech. Virtual-office automatic noise removal is separate and does not apply to live/offline translation or device-audio input.