CEVA Introduces ClearVox Solution for Enhanced Speech Intelligibility in Voice-Enabled Devices

January 8 2018, 03:00
CEVA, a leading licensor of signal processing platforms for smart, connected devices, introduced ClearVox, a new software suite of advanced voice input processing algorithms aimed at enhancing speech intelligibility and voice clarity for voice-enabled devices. ClearVox is licensed exclusively for the CEVA-TeakLite-4 and CEVA-X2 audio/voice DSPs, and extends the company’s voice portfolio with multi-mic beamforming, speaker direction of arrival, noise suppression and echo cancellation algorithms.

ClearVox is basically a software suite of advanced voice input processing algorithms for CEVA sound DSPs: CEVA-TeakLite-4 (ultra-low-power) and CEVA-X2 (high-performance), providing voice pickup front-end processing for the emerging voice activation and voice interaction market, such as smart speakers, advanced headsets, and IoT devices. ClearVox incorporates multiple algorithms such as beamforming, acoustic echo cancellation, and noise suppression, based upon CEVA’s microphone array technology, and delivered in configurations for near-field applications like headsets and wearables, or far-field applications like smart speakers, infotainment and advanced voice-enabled devices.

As voice becomes the primary user interface for cloud-based assistant services, a clean voice signal in any given environment is mandatory to ensure accuracy and intelligibility by conversational assistants such as Alexa, Baidu's DuerOS, Bixby, Cortana, Google Assistant and Siri. Headsets, smart speakers, mobile phones, wearables, automotive infotainment and home security systems are just some of the sizeable markets that vendors and chipmakers are targeting with voice-enabled services. However, they face significant technology barriers to incorporate and configure complicated algorithms such as multi-mic beamforming, multi-channel acoustic echo cancellation and noise suppression in their products. 

To reduce the significant barriers-to-entry for these markets, CEVA has introduced the ClearVox voice input software package. Developed in-house and leveraging CEVA's vast expertize in audio and voice processing, ClearVox incorporates advanced algorithms that cope with different acoustic scenarios and microphone configurations, including optimized software for speaker direction of arrival, multi-mic beamforming, noise suppression and acoustic echo cancellation, as well as the related firmware and driver software. It is offered in two pre-packaged software form factors: one for near-field applications including headsets and wearables, and the other for far-field applications targeting the emerging smart speaker, infotainment and other advanced voice-enabled applications. The accompanying SDKit is available for customers to configure and adopt ClearVox to their unique product specifications, such as the number and geometry of microphones, echo tail length and more.
CEVA ClearVox Reference Design available on CEVA's development platform - a 500 MHz audio-and-sensing DSP running the DSP RTOS platform, which includes a multi-microphone daughter board with 13 microphones, and allows developers to measure power and optimize their DSP firmware. ClearVox is implemented on input path, using from 1 to 8 mics, with audio playback–decode and post-processing.

Erez Bar-Niv, Chief Technology Officer at CEVA, commented: "Voice input technologies are some of the most complicated and diverse algorithms to design, yet have become a 'must-have' for every voice-enabled device. Our ClearVox technology takes advantage of the latest academic research in algorithms for beamforming and signal direction of arrival to achieve a truly robust and exceptionally clear voice signal, regardless of background noise, acoustics, environment or product design. Our customers can now benefit from a single-source offering for the complete voice input solution, both the DSP and the software reducing design complexity and their time-to-market."

ClearVox is available for general licensing in Q2 2018.
related items