Multilingual speech understanding for intelligent edge devices

Multilingual speech understanding for intelligent edge devices

Market news |
By Rich Pell

CEVA, Inc., and have partnered to offer ultra-low power speech-to-intent products for intelligent edge devices. A suite of speech-to-intent technologies from has been ported and optimized for CEVA’s low power audio and sensor hub DSPs, providing a high performance, robust implementation for OEMs and ODMs looking to integrate intelligent voice activation and control into their wearables, consumer devices and IoT products. provides embedded, noise robust and multilingual speech understanding systems capable of running fully offline on small footprint and low power devices. technology is private-by-design and can support any language and accent, enabling users to speak to their devices in their native language, naturally, and without sacrificing their privacy.

CEVA’s powerful audio and sensor hub DSPs, including the CEVA-X2, CEVA-BX1, CEVA-BX2 and SensPro family, enable the full suite of speech-to-intent technologies to run seamlessly in always-on mode. These DSPs can also run other software and algorithms that further enhance the performance and feature set, including ClearVox front-end noise reduction, MotionEngine for sensor fusion and the SenslinQ framework for contextual awareness.

“CEVA’s leadership position in licensable low power DSPs and software for voice-enabled embedded devices makes them an ideal partner to further proliferate our speech-to-intent software in the market,” said Vikrant Tomar, Founder and CTO, “Voice activation and control is emerging as one of the most sought-after technologies in an increasingly contactless world, and together we’re bringing a cost-effective and highly accurate edge AI that can understand intent from speech, even in the noisiest environments.”

“’s speech-to-intent technology with multilanguage support running on our DSPs is ideal for power-constrained intelligent devices where voice is the primary user interface,” said Moshe Sheier, Vice President of Marketing at CEVA. “Having all the speech processing take place on the edge device ensures privacy of the data, low latency and instantaneous response. Together, we are lowering the entry barriers for adding high quality, naturally spoken voice control to any device.” utilizes a unique approach to multi-language speech understanding. Instead of the traditional Cloud-based approach of transcribing speech to text and then using natural language processing to extract meaning, the company has developed an end-to-end spoken language understanding technology that directly extracts intent from the input speech alone. This approach allows to design speech understanding models that are much smaller in size, yet provide high accuracy even in noisy environments. systems are capable of recognizing up to 1000s of intents in a small model size of 100s of KBs. Furthermore,’s ability to build multiple languages into a single model means that users can switch seamlessly between languages when interacting with their device, without the need to configure language settings in between. These advantages make ideal for smart devices in the home and office, consumer wearables and factory robot automation.

CEVA’s scalable audio and sensor hub DSPs are optimized for sound processing applications ranging from always-on voice control up to multiple sensors fusion. They have been specifically designed to tackle multi-microphone speech processing use-cases, high quality audio playback and post-processing, and on-device sound neural network implementations. In addition, a large 3rd party ecosystem of audio/voice software, hardware and development tools companies have optimized their systems for CEVA DSPs, for a wide array of use cases and applications.

If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :    eeNews on Google News


Linked Articles