Free (On-Device) Setup Tutorial

1Select "Free" as Provider

Open Sokuji and navigate to the Settings panel. Select "Free" as your AI provider from the provider dropdown.

2Download an ASR Model

Choose an ASR model for your source language. Sokuji offers 42 offline/streaming models powered by sherpa-onnx WASM and WebGPU (Cohere Transcribe, Voxtral Mini, Whisper), covering 99+ languages. Models range from lightweight (~5MB) to high-accuracy (~2GB+). We recommend Cohere Transcribe or Voxtral Mini 4B Realtime for the best accuracy and speed.

3Download a Translation Model

Select a translation language pair. Choose TranslateGemma for the best quality and speed, Opus-MT for fast CPU-based translation, or Qwen LLMs via WebGPU for multilingual translation with higher quality.

4Download a TTS Model

Download a TTS model for spoken output. 136 models are available across 52 languages, powered by Piper, Coqui, Mimic3, and Matcha engines. If Translation I hear is on in the Speech block (Other or Both mode), the other side's translation is spoken to you too: its direction gets its own TTS slot, with its own model and voice. Each direction keeps its own voice.

5Start Translating

Click the Start Session button. Recognition and translation happen on your device in real time; with an Edge TTS voice selected, speaking the translation needs a network connection.

What is the Free provider?

Free is a fully on-device AI translation pipeline that runs ASR, translation, and TTS without any cloud services:

• Complete privacy — all audio and text processing stays on your device
• Works offline after initial model download
• No API keys, accounts, or subscriptions required
• Powered by WebAssembly (WASM) and WebGPU for native-like performance
• Free to use with no usage limits

Available Models

The Free provider supports a wide range of models for each stage of the translation pipeline:

• ASR: 42 models including Cohere Transcribe, Voxtral Mini 4B Realtime, Whisper, SenseVoice, Moonshine, NeMo, and Zipformer — covering 99+ languages
• Translation: 74 Opus-MT language pairs for fast CPU translation, plus TranslateGemma and Qwen LLMs via WebGPU for multilingual support
• TTS: 136 models across 52 languages using Piper, Coqui, Mimic3, and Matcha engines

VAD Settings (speech detection)

With Speech Mode set to Auto in Settings, click VAD Settings next to it to tune the on-device speech detection. In Push-to-Talk and Push-to-Translate you control where each sentence starts and ends yourself:

• Max Speech Duration: how long one utterance can run before it is split (10–40 s, default 30 s); some models split sooner, at the longest stretch they transcribe correctly
• Pre-Speech Padding: keeps a little audio from just before speech is detected (up to 2 s, default 0.8 s), so the first sound of a sentence is not cut off; shown only for the models that support it
• Smart (Smart Turn): instead of always waiting out a fixed silence, checks whether you have finished speaking and ends the sentence about 0.4 s after you stop; a pause mid-sentence keeps it listening. Turning it on downloads a 31 MB model once; it works with the WebGPU speech recognition models
• Restore defaults: puts these settings back to their defaults

Hardware Requirements

• CPU (WASM): Works on any modern browser — no special hardware needed
• WebGPU: Optional GPU acceleration for Cohere Transcribe, Voxtral Mini, Whisper ASR, TranslateGemma, and Qwen translation models — significantly faster with a dedicated GPU
• Recommended: 4GB+ RAM for comfortable model loading
• Storage: Models range from 5MB to 200MB+ depending on accuracy level
• Modern browser required: Chrome 113+, Edge 113+, or Firefox 120+ for WebGPU support

TTS Settings

• Speech Speed — Controls how fast the synthesized speech is played. Adjust the speed slider to make the output faster or slower to suit your preference.
• Speaker ID — Controls the voice timbre of the synthesized speech. Many models support multiple speaker IDs, each with a different voice. Some multi-speaker models include voices of different genders, so you can choose a male or female voice by selecting different speaker IDs. The Speaker ID, or the Edge TTS voice, is kept per direction: what you pick in one direction's TTS slot does not change the other direction.

Common Issues

Voice reset after updating: Since Sokuji 0.44.0 each direction keeps its own voice. A voice picked in an earlier version was shared by both directions, so after the update each direction starts once from its default voice. Pick the voice again in each direction's TTS slot.

Translation I hear is text only: If the warning "Translation I hear is text only: not enough memory to load the other side's speech model too." appears, the models of both directions do not fit in memory together, so the other side's translation starts as text only while yours is still spoken. Its Settings button opens that direction's TTS slot: pick a smaller model, or Edge TTS, which takes no local memory.

Edge TTS stopped speaking in the browser extension: Edge TTS now synthesizes speech on api.msedgeservices.com. When the extension updates to Sokuji 0.44.0, Chrome or Edge asks you to allow access to that site and keeps the extension disabled until you accept: open the browser's Extensions page and accept the new permission. The desktop app needs nothing.

Model download fails: Check your internet connection for the initial download. Models are cached locally in IndexedDB, so subsequent uses work offline. Try clearing browser data if downloads are corrupted.

ASR not recognizing speech: Ensure you selected the correct ASR model for your source language. Try a larger model for better accuracy. Check that your microphone is working and permissions are granted.

Translation quality is low: Opus-MT models are optimized for speed over quality. For better results, try TranslateGemma or Qwen LLM models if your device supports WebGPU. Ensure you selected the correct language pair.

WebGPU not available: WebGPU requires a compatible browser and GPU. Fall back to WASM-based models (sherpa-onnx, Opus-MT) which work on any modern browser without GPU requirements.

Performance Tips

1. Use WebGPU models when available for significantly faster processing
2. Close other browser tabs to free up memory for model loading
3. Choose smaller models for low-end devices or when speed is prioritized
4. WASM models (sherpa-onnx, Opus-MT) are more compatible but slower than WebGPU alternatives
5. Monitor browser memory usage — large models may require 2GB+ of browser memory