Supported AI Providers
Sokuji supports multiple AI providers for real-time speech translation. Each provider offers different capabilities, models, and pricing structures. Free mode requires no API key and runs on your device at no cost; only some voices need an internet connection.
Setup Instructions
To use a cloud AI provider, obtain an API key from the provider's website and configure it in Sokuji's settings panel. Free mode requires no API key — just download the models and start translating.
OpenAI
- GPT-4o Realtime Preview models
- 8 premium voice options
- Advanced turn detection modes
- Built-in noise reduction
- 60+ languages supported
- Template mode for custom prompts
Google Gemini
- Gemini 2.0 Flash Live models
- 30 unique voice personalities
- Automatic turn detection
- 35+ languages with regional variants
- Built-in transcription
- High token limits (8192)
PalabraAI
- Real-time WebRTC translation
- 60+ source languages
- 40+ target languages
- Low latency streaming
- Automatic audio processing
- Specialized for live translation
Doubao AST 2.0
- End-to-end speech-to-speech translation
- Voice cloning, or a fixed voice from a library of 500+
- Low-latency simultaneous interpreting
- 20 languages plus Cantonese and Wu, 8 of them spoken
- App ID + Access Token, or an API key
- ByteDance cloud infrastructure
Soniox
- Real-time speech-to-speech translation
- Transcribe and translate 60+ languages in one API call
- Automatic language detection
- 200 built-in voices and voice cloning
- Spoken translation or subtitles only
- Face-to-face conversation and speaker labels
- Soniox cloud infrastructure
Free
- Fully offline — no API key or internet needed
- 40+ ASR models covering 99+ languages
- 55+ translation language pairs + Qwen LLMs
- 136 TTS models across 53 languages
- CPU (WASM) and WebGPU acceleration
- Privacy-first — all data stays on-device
Choosing a Provider
OpenAI: Best for high-quality voice synthesis and advanced features
Gemini: Great for multilingual support and automatic processing
PalabraAI: Optimized for real-time translation with minimal latency
Doubao AST 2.0: Doubao speech-to-speech simultaneous interpretation that clones the speaker’s voice or speaks in a fixed voice of your choice
Soniox: Real-time speech-to-speech translation across 60+ languages with automatic language detection
Local Inference: Run everything on your device with no data leaving your machine — perfect for privacy-sensitive use cases
Need Help?
For setup guides, troubleshooting, and provider comparisons, visit our GitHub repository. GitHub Repository