Knowles Announces The World’s First Context-Aware, Machine Learning Optimized Audio Processor For Complex Voice Interactions

October 17 2017, 04:10
Knowles announced a new audio processor platform, the IA8508 open DSP platform, with four high performance, power efficient, heterogeneous cores and 5.7MB of shared memory for a variety of new use cases. The IA8508 advanced audio processor is optimized for performance and power in both near and far-field voice applications – making it the perfect solution for mobile phones, ear-worn products, and IoT devices such as digital assistants, smart speakers and TVs.

The IA8508 platform addresses the performance limitations of today's most popular voice-enabled devices by improving recognition especially in far-field and high noise environments, enabling context awareness, and increasing local processing capability. Users are now able to move audio processing to the edge of the network – embedding intelligence into the device where it is closer to the user and the environment.

Furthermore, the IA8508 is designed for Machine Learning (ML) through Deep Neural Network (DNN) hardware acceleration, optimized architecture and proprietary instruction set. As a result, the new Knowles’ audio processor can perform specific machine learning tasks 10 times more efficiently than existing products. In addition, the IA8508 leverages light weight floating point operations, enabling ease of programmability, high dynamic range and low power comparable to fixed point operations.

“The market opportunity for voice-powered devices is massive. From mobile phones to ear-worn devices, and smart speakers to TVs and appliances, voice-powered interactions have emerged as a critical and necessary feature,” says Mike Polacek, President of Intelligent Audio at Knowles. “People are looking to engage with technology through natural, spoken commands, across mobile, ear-worn and IoT products, and are racing to develop and deploy the technology to enable it. With core capabilities in acoustics, digital signal processing and algorithms, Knowles’ intelligent audio expertise is unparalleled and places the company in a perfect position to solve the issues hampering voice technology.”

Knowles' unique combination of four heterogeneous processors sets the IA8508 apart in 4 major ways. First, to solve issues related to noise and distance, device makers need the ability to combine a multi-microphone array with large memory, powerful computing, and sophisticated algorithm features including acoustic echo cancellation (AEC), dynamic beam forming, steering, sound classification, and noise suppression. The IA8508 supports an array of up to eight microphones and is equipped with four heterogeneous cores capable of simultaneously running high performance algorithms leveraging the 5.7MB of memory. The Delta-Max core (DMAX) with specialized instruction set serves as the high-performance core.

Second, latency. Many audio applications require low-latency real-time processing, such as active noise cancellation and asynchronous sample rate conversion for mixing of various different audio signal sources. The IA8508 has an optimized single sample processor (SSP) core, guaranteeing low latency.  

Third, power consumption. Voice user-interfaces require always-listening devices. The IA8508 features a low power core to listen to a variety of key phrases concurrently. In addition, complex audio processing and ML run more efficiently due to the optimized instruction set. The Hemi-Delta core (HMD) is tailored to guarantee low power dynamic processing; in addition, proprietary memory architecture and dynamic voltage scaling guarantee low stand-by power consumption.

And fourth, integration. While the above DSP cores are optimized for performance, latency, and power efficiency, the IA8508 also includes an industry standard ARM Cortex M4 core for general purpose controller capability, allowing for immediate porting of existing legacy solutions.

Also announced today is the Knowles open DSP partner program, which brings together leading contributors to the intelligent voice ecosystem for improving audio performance in a variety of use-cases. Earlier this year, Knowles announced the world's first smart mic with an embedded open DSP, the IA-610. Since then, the company has continued to add world-class partners to leverage its open DSP platforms. The company plans to continue expanding this program, collaborating with algorithm and cloud partners to develop and run code on its processors for solving complex audio problems.

“Knowles continues to innovate, most recently through its open DSP platform architecture providing both power efficiency and versatility,” says Todd Mozer, CEO of Sensory. "The Knowles open DSP platform environment enables quick and smooth integration of 3rd-party IP and this is evidenced by the recent successful enablement of Sensory's TrulyHandsfree technology.”

The combined strength of the IA8508 and the open DSP partner program demonstrates Knowles' leadership across the full spectrum of the voice interaction experience, from smart mics to audio processors. Coupled with its ecosystem of algorithm and cloud partners, Knowles is poised to enable mobile, earn-worn and IoT customers to capitalize on growing consumer demand for voice applications.

“Knowles' market leadership in microphones combined with our technology advancement in digital signal processing and algorithms uniquely positions us to understand and solve complex audio problems,” adds Jeffrey Niew, President and CEO of Knowles. "With our growing line-up of open audio DSPs, Knowles is enabling our customers and partners to push the boundaries of audio processing and create powerful new user experiences."
related items