Reka Multilingual Speech
Translate and transcribe real-time audio in multiple languages
A state-of-the-art speech-to-text and speech-to-speech transcription and translation model, especially suitable for deployment on devices (e.g., cellphones, wearable devices, cars, robots) and for low-latency applications.
Excels at low-latency and accurate conversational speech-to-speech translation model, deployable on devices
Multilingual support
To and from 10 languages: English, Japanese, Korean, Chinese, Arabic, French, Spanish, German, Italian, Portuguese.
Outperforms popular open source models such as Whisper-L v3 which is three times larger on transcription and translation quality.
State-of-the-art performance
Built for on-device deployment
It has ~900 million parameters and can be deployed on devices with less than 1GB memory size. This model can be customized further. Contact us for support.
