XMOS has launched VocalFusion XVF3620, a next-generation voice processor that combines AI-powered noise reduction with a complete on-chip voice processing pipeline to deliver reliable voice capture in noisy, reverberant, and dynamic environments. The new platform combines AI denoising features, full-duplex acoustic echo cancellation with residual filtering, two-microphone beamforming, and automatic gain control. A solution designed for consumer electronics and any applications that require robust, real-world voice performance.
XMOS has launched VocalFusion XVF3620, a next-generation voice processor that combines AI-powered noise reduction with a complete on-chip voice processing pipeline to deliver reliable voice capture in noisy, reverberant, and dynamic environments. The new platform combines AI denoising features, full-duplex acoustic echo cancellation with residual filtering, two-microphone beamforming, and automatic gain control. A solution designed for consumer electronics and any applications that require robust, real-world voice performance.
As a specialist in audio, voice, and physical AI SoCs, XMOS understands the fast-expanding requirements from manufacturers to deliver reliable and advanced voice interfaces in a growing class of products. The exciting possibilities with AI-powered audio processing have boosted expectations from product managers that require a new approach. The new VocalFusion XVF3620 processor has been developed directly to address accurate speech capture and natural communication in real-world conditions.
By combining AI denoising, fast-adapting acoustic echo cancellation (AEC), two-microphone beamforming, and automatic gain control (AGC), the device enables high-quality voice interaction across a wide range of consumer and industrial applications. During the development stage, XMOS tested this new solution against a wide range of real-world noise environments, including traffic, rain, and thunderstorms, air conditioning, and busy hospitality environments, with consistently high speech quality, strong perceptual performance, and significantly improved signal clarity across all tested noise levels.
Leveraging its XCORE platform, XMOS created a processor that excels in real-time, parallel processing, ready to meet the biggest challenges facing voice-enabled products. XCORE offers a deterministic, parallel architecture ideal to perform multiple tasks at the same time without interference, and complete them reliably and predictably across all deployment scenarios.
Designed for smart home appliances, service robotics, industrial assistants, and outdoor installations, the XVF3620 suppresses background noise while preserving speech clarity. Its AI-based denoising technology can be tuned to specific application requirements and speech recognition engines, helping optimize performance for both human communication and voice AI interfaces. The model separates speech from background noise and was trained on a wide range of noise types, and trained to remove reverb, in order to preserve speech quality. Its low latency makes it suitable for intercom and control applications.
The processor incorporates full-duplex acoustic echo cancellation with advanced residual echo filtering, enabling reliable barge-in performance, even while audio is being played back. To simplify product development, the XVF3620 also integrates a complete digital voice processing chain on a single device, including PDM-to-PCM conversion, beamforming, AI denoising, multiband compression, equalization, and AGC. This reduces system complexity, lowers costs, and accelerates time-to-market for manufacturers developing voice-enabled products.
“Voice is becoming the primary interface between people and technology, but noisy and cluttered real-world environments still pose a significant challenge for traditional voice systems. The XVF3620 combines cutting-edge AI audio processing with our proven XMOS voice technology to bring natural and reliable speech capture wherever the products are deployed,” says Tim Robjohns, Director of Product Marketing at XMOS.
XK Voice L71 development kit.
The VocalFusion XVF3620 processor offers audio interface options via USB or I²S (master or slave), and control interfaces via I²C, UART, or USB HID. It supports a range of microphone configurations for omnidirectional capture or spatial enhancement for noisy environments and offers direct connection of PDM microphones to XCORE, and an integrated far-end DSP for EQ, limiting, and multiband dynamics to achieve the best performance from any loudspeaker.
The XK Voice L71, 2-Mic voice development kit, is available to help evaluate the XVF3620 and prototype voice-enabled products. It supports USB plugin accessory implementations (with a USB Control Interface) and also built-in voice implementations (with an I2C Control Interface).
www.xmos.com
Functional block diagram of XVF3620 in UA (USB) configuration.