HiFi 5 DSP delivers up to 4X improvement on neural network-based speech recognition algorithms
SAN JOSE, Calif. — (BUSINESS WIRE) — October 31, 2018 — Cadence Design Systems, Inc. (NASDAQ: CDNS) today announced the Cadence® Tensilica® HiFi 5 DSP for audio and voice, the first IP core optimized for high-performance far-field processing and artificial intelligence (AI)-based speech recognition processing. This fifth-generation HiFi DSP offers 2X audio processing and 4X neural network (NN) processing improvements versus the HiFi 4 DSP, making it ideal for voice-controlled user interfaces in digital home assistants and automotive infotainment.
This press release features multimedia. View the full release here: https://www.businesswire.com/news/home/20181031005165/en/
The Cadence Tensilica HiFi 5 DSP for audio and voice is the first IP core optimized for high-performance far-field processing and artificial intelligence (AI)-based speech recognition processing. (Graphic: Business Wire)
For more information, visit http://www.cadence.com/go/hifi5.
Following the popularity of digital home assistants, voice-controlled user interfaces are increasingly important to manufacturers of innovative consumer products. Advanced DSP algorithms are rapidly evolving to eliminate noise and isolate the speakers’ voice for increased understanding, requiring higher processing capabilities and improved energy efficiency. Additionally, NN-based speech recognition algorithms are performing more tasks locally, rather than in the cloud, due to concerns of latency, privacy and network availability.
“To meet the extremely difficult challenge of bringing computationally intensive NN-based far-field processing and speech recognition algorithms to energy-sensitive devices, Ambiq Micro chose to be the first silicon licensee of Cadence’s HiFi 5 DSP,” said Aaron Grassian, vice president of marketing, Ambiq Micro. “Porting the HiFi 5 DSP to Ambiq Micro’s Subthreshold Power Optimized Technology (SPOT) platform enables product designers, ODMs and OEMs to take the most advantage of technology from audio software leaders like DSP Concepts and Sensory by adding voice assistant integration, command and control, and conversational UIs to portable, mobile products without sacrificing quality or battery life.”
Key features of the HiFi 5 DSP include:
- Five very long instruction word (VLIW)-slot architecture capable of issuing two 128-bit loads per cycle
2X MAC capability versus HiFi 4 DSP for pre- and post-processing,
- Support for eight 32x32-bit or 16 16x16-bit MACs per cycle
- Optional eight single-precision floating-point MACs per cycle
4X MAC capability versus HiFi 4 DSP for NN processing, includes:
- 32 16x8 or 16x4 MACs per cycle
- Optional 16 half-precision floating-point MACs per cycle
- The new HiFi NN library offers a highly optimized set of library functions commonly used in NN processing (especially speech). These library functions can easily be integrated into popular machine learning frameworks.
- Software compatibility with the complete HiFi product line totaling over 300 HiFi-optimized audio and voice codecs and audio enhancement software packages.
“The dramatic rise in popularity of digital home assistants that feature compelling voice UI experiences has sparked a new wave of innovation in far-field processing algorithms and in neural network-based speech recognition,” stated Larry Przywara, group director of marketing, Tensilica audio/voice IP at Cadence. “The HiFi 5 DSP addresses both of these high-performance compute concerns with improved floating- and fixed-point DSP capabilities and native support for new data types, saving memory and power. The HiFi 5 DSP maintains code compatibility with the existing HiFi DSP product line, which means a large ecosystem of audio software is already available.”
“AI processing at the edge is a reality in today’s digital home appliance market,” stated Mike Demler, Senior Analyst at The Linley Group. “Due to consumer demands for lower latency, increased privacy and more natural voice UI interactions, the processing work load on device is increasing rapidly. The HiFi 5 DSP offers the performance needed to handle front-end processing including echo cancellation and noise-reduction tasks, and its new 32-MAC/cycle neural-network engine with native support for smaller weight sizes offers an efficient solution for running complex speech-recognition algorithms.”
“An interesting challenge for OEMs is the demand to incorporate new
DSP-intensive audio features, while at the same time supporting shorter
and shorter development schedules,” said Paul Beckmann, Chief Technology
Officer, DSP Concepts. “To meet both of those competing demands, product
makers can choose powerful processing cores like the HiFi 5 DSP with its
2X improvement in floating-point performance, and then rapidly build out
their audio features with graphical configuration tools like Audio
Weaver Designer. Without both capabilities—modern silicon and modern
development tools—it’s really tough to keep pace.”