Why dedicated hardware for HMI processing?
Because of the memory and computing bandwidth constraints of HMIs (such as voice recognition), which traditionally leverage a shared CPU resource, compromises have to be made that affect the quality of the end user experience.
For example, in voice recognition, designers have had to make tradeoffs between speed and accuracy on shared-resource embedded systems. Larger acoustic models enable better accuracy but require more processing power to avoid unacceptable latencies in responses. And, as users begin to expect more of their voice processing interfaces -- such as distinguishing between genders, noises, dialects, accents, and multiple languages -- these feature-rich acoustic models will grow exponentially in size over time, and reliable fast-access memory will become more critical for increased performance. Unfortunately, today's resource-shared, resource-constrained hardware platforms can't provide acceptable processing power for today's largest acoustic models. The industry has had to compromise by developing compact acoustic models that provide minimum acceptable levels of accuracy within the minimum acceptable response time.
A simple solution to this has been recently introduced to the design industry -- a dedicated hardware coprocessor that can be used to accelerate the processing power and speed of certain phases of the voice recognition process. The first of these, the Spansion Acoustic Coprocessor, addresses the acoustic scoring phase of voice recognition, offloads CPU time, and reduces latency by up to 50 percent. Additionally, the Spansion Acoustic Coprocessor offers enough processing power for today's largest acoustic models, as well as acoustic models up to 10 times the size, eliminating the latency-accuracy compromises needed when designing embedded voice recognition solutions on shared-resource platforms.
These types of dedicated UI coprocessors leverage advanced flash memory technology to provide instant-on response, high reliability, and performance across a range of applications.
How will advanced HMIs look?
Though voice recognition represents the current wave of HMIs in embedded systems, we are only at the beginning of providing truly advanced HMIs that incorporate things such as natural language understanding, image recognition, or emotional awareness. However, recent advances in flash memory have driven the industry to new and innovative heights. Specialized hardware, equipped with advanced memory technology and logic, will enable richer and improved software models to bring us closer to these advanced HMIs. And as flash memory innovation continues to advance, so will our capability to design richer and more natural interfaces that ultimately improve the user experience.
By challenging us to keep innovating products with new architectures to support the best customer experience, users have ensured that innovation in flash memory is moving forward at high speed. The next frontier and challenge in providing a rich user experience is here -- voice recognition powered by advanced flash memory technologies and dedicated hardware.
Rik Graulus is product line director for Spansion.