LAPLACE Chatterbox's simultaneous interpretation is built on the Soniox real-time speech recognition API. It transcribes speech to text while you stream and sends it automatically to the current live room's chat as danmaku
It supports multilingual recognition, real-time translation, and automatic segmented sending — suited to interpreting spoken content into a live room as it happens
Features
- Built on Soniox's newest real-time recognition model, with low latency and stable results on mixed-language Chinese, Japanese, and English speech
- Language hints let you declare which languages may appear (中文, English, 日本語, 한국어), raising accuracy on mixed-language speech
- Results past the configured character count are split automatically, preferring sentence-ending punctuation (
。？！…), then clause punctuation (，、；：), and only then a space or the character limit, so every danmaku stays semantically whole - Interpretation text can be wrapped in 【】, so viewers can tell interpretation danmaku apart from ordinary ones in chat
- Real-time translation, once enabled, converts each result into the target language and sends the translation instead of the original
- A failed send can optionally trigger an AI evasion retry, which detects sensitive words and tries to slip past the filter
- Interpretation sends go through the shared send queue, ranking above Auto Send and Auto Blend but below manual sends
- The API key is stored only in your local browser, never uploaded to any server
Usage
Setup
- Register at the Soniox console and get an API key
- In the Chatterbox panel, switch to the Simultaneous Interpretation tab and enter the API key under Soniox API Settings
Start Interpreting
- Switch to the Simultaneous Interpretation tab
- Check the language hints you need — only the languages that will actually come up
- Click Start Interpretation, then allow the microphone permission request your browser raises
- Results appear in the panel as they arrive and go to chat once a pause in speech is detected
Real-Time Translation
- Check Enable Real-Time Translation under Real-Time Translation Settings
- Pick a target language (English / 中文 / 日本語)
- With it on, what gets sent is the translation, not the original recognized text
Options
| Option | Description |
|---|---|
| Language Hints | Check the languages that may appear; multiple selections are allowed, which raises recognition accuracy on mixed-language speech |
| Auto-Split Length | Splits into multiple danmaku once the text runs past this many characters, 40 by default |
| Send Danmaku Automatically After Recognition | Unchecked, results are not sent automatically and the transcript only shows in the panel |
| Wrap Interpretation Text in 【】 | Wraps every danmaku in 【】, reserving 2 characters of the auto-split length for the brackets |
| Enable Real-Time Translation | Sends the translation rather than the original |
| Translation Target Language | English / 中文 / 日本語 |
Simultaneous interpretation reads the browser's microphone input. To transcribe system audio such as in-game voice or another app's output, route that audio to a virtual microphone device at the OS level
Notes
- The Soniox API bills by usage — see the Soniox pricing page
- Results are flushed and sent automatically within about 5 seconds of a detected pause in speech; stopping interpretation triggers one final send of whatever remains
- Interpretation danmaku are sent at a minimum interval of 1.1 seconds, staying clear of bilibili's rate limit
Last updated on September 19, 2026