Smart Edge SoCs Blend Neural Net Acceleration and Far-Field Voice

Synaptics has announced its new AS3xx series, part of its Smart Edge AudioSmart family. The series includes fully integrated SoCs comprising neural network acceleration, a proprietary wake word engine with support for custom wake words and advanced far-field voice processing. Designed using a 22 nm process node, and targeting a broad variety of voice-enabled Smart Home devices including hubs, Wi-Fi repeaters, speakers and appliances, the AudioSmart AS371 is sampling now and incorporates a new machine learning engine using innovative SyNAP (Synaptics Neural Network Acceleration and Processing) technology.
According to Synaptics, Intelligence at the edge results in an enhanced user experience through better response times and robustness. For example, performing Automated Speech Recognition (ASR) and Natural Language Understanding (NLU) locally on-device ensures core Smart Home voice control works when there is poor or intermittent internet connectivity. SyNAP enables advanced features such as user identification and behavioral prediction which allows voice assistants to perform ambient computing and a more intuitive interaction with users.

The AS371 solution with SyNAP on-device intelligence enhances consumer privacy by greatly reducing the need to continuously send personal data to the cloud. SyNAP also takes the right steps to support GDPR compliance as well as other current and emerging regulations. The AudioSmart AS3xx family introduces the integrated and purpose-built far-field voice processing and wake word technology developed by Synaptics. Having been used in a variety of global retail products for over seven years, Synaptics continues to improve its voice processing solution to further improve voice pick-up in noisy, real-world conditions and barge-in capabilities during very loud playback.

Through in-house design of the SoC, wake word engine, voice pre-processing and playback enhancement, Synaptics has optimized the complete system to achieve the highest possible far-field voice performance at affordable consumer electronics price points. Furthermore, by providing the complete solution including acoustics system expertise, Synaptics greatly reduces the time-to-market for Smart Home product manufacturers and allows them to add an excellent voice-first experience while maintaining focus on their core expertise. In addition, Synaptics’ AS3xx solutions are fully compatible with all voice assistant platforms worldwide.

New AS3xx Products:

  • AS371 (for voice-enabled devices with SyNAP) Sampling Now
  • AS390 (for voice-enabled devices with displays) Target sampling CY19Q1
  • AS350 (for voice-enabled low-power devices) Target sampling CY19Q1
  • AS320 (for voice-enabled microcontroller-based devices) Target sampling CY19Q1

Synaptics | www.synaptics.com

 

Tensilica HiFi 3z DSP IP Core Provides Enhanced Voice and Audio Processing

Cadence Design Systems has announced the Cadence Tensilica HiFi 3z DSP IP core for system-on-chip (SoC) designs targeted for the latest mobile and home entertainment applications, including smartphones, augmented reality (AR)/3D goggles, digital TVs and set-top boxes (STBs). The new HiFi 3z architecture offers more than 1.3X better voice and audio processing performance than its predecessor, the HiFi 3 DSP, which leads the industry in the number of audio DSP cores shipped.

20170727190416_Cadence-Tensilica-HiFi-3z-DSP-IPFrontWeb

Higher voice sample rates require more complex voice pre-processing. Enhanced Voice Services (EVS), the latest mobile voice codec supporting voice over LTE (VoLTE), supports up to a 48kHz sample rate, compared to 16kHz for the previous AMR-WB codec. The new HiFi 3z DSP delivers more than 1.3X better performance for EVS than the HiFi 3 DSP core. Home entertainment is driving a similar workload increase as audio codecs like Dolby AC-4 and MPEG-H transition from channel-based to object-based. In addition, audio post-processing functions such as Waves Nx 3D/AR audio and the immersive audio of Dolby Atmos-enabled TVs are driving higher complexity signal processing. The HiFi 3z DSP provides more than 1.4X better performance on Dolby Atmos-enabled TVs than the HiFi 3 DSP.

The HiFi 3z DSP offers a number of architecture and instruction set architecture (ISA) improvements over the earlier HiFi 3 DSP, including:

– Dual load/store

– Advanced FLIX bundling (multiple-base ISA operations per cycle)

– Double the MACs for 16×16 (octal MAC)

– Enhanced ISA for accelerating FFTs, FIRs and IIRs

– New instruction extensions to improve codec (especially EVS) performance for mobile

– 4-way, 8-bit load for improved voice trigger performance

– 8-way, 8-bit load for reduced neural network memory usage

Tensilica HiFi DSPs are the most widely licensed audio/voice/speech processors, with support for over 200 proven software packages and more than 95 software partners in the Tensilica Xtensions partner program. More than 75 top-tier semiconductor companies and system OEMs have selected Tensilica HiFi DSPs for their audio, voice and speech products.

Cadence | www.cadence.com/go/hifi3z

SoC FPGA Development Kit for Audio & Processing Applications

Coveloz recently announced the availability of its Pro Audio Ethernet AVB FPGA Development Kit, which is a ready-to-play platform for building scalable, cost-effective networked audio and processing applications built on modular hardware.Covelozpro-audio-dev-kit

Coveloz introduced its Networked Pro Audio SoC FPGA Development Kit during the Integrated System Europe (ISE) show in Amsterdam. According to the company, the new platform will enable manufacturers to achieve faster AVnu certification for new AVB solutions, creating an ideal development environment for live sound, conferencing systems, public address, audio post production, music creation, automotive infotainment and ADAS applications.

At the heart of the Coveloz development platform is a highly integrated System-on-Module (SOM), featuring an Altera Cyclone V SoC FPGA, which includes a dual-core ARM A9 processor, DDR3 memory and a large FPGA fabric, all in a low cost and compact package. The kit includes a multitude of networking and audio interfaces, including three Gigabit Ethernet ports as well as I2S, AES10/MADI, AES3/EBU and TDM audio.

Coveloz provides FPGA and Linux firmware enabling designers to quickly build AVnu Certified products for the broadcast, pro-audio/video and automotive markets. The platform is aimed at time-synchronized networks and includes grandmaster, PPS and word clock inputs and outputs as well as high quality timing references.

The Coveloz development kit is also host to the BACH-SOC platform, which integrates AES67 and Ethernet AVB audio networking and processing. Both SoC and PCIe-based FPGA implementations are available.

The Coveloz Bach Module is a full-featured and programmable audio networking and processing solution for easily integrating industry-standard AES67 and/or Ethernet AVB/TSN networking into audio/video distribution and processing products. The solution enables products with over 128+128 channels of digital streaming and 32-bit audio processing at 48, 96, or 192 kHz.

Supporting a wide range of interfaces, Coveloz complements the development platform with a comprehensive software toolkit and engineering services to help manufacturers reducing time to market. Coveloz also provides application examples to demonstrate the capabilities of the BACH-SOC platform.

The programmable BACH-SOC can be customized to a particular application in many ways—for instance, from selecting the number and type of audio interface to choosing audio processing alone, transport alone, or a combination.

Source: Coveloz