Archive

Posts Tagged ‘artificial intelligence’

Amlogic A113X1 6-Mic Far-Field Devkit is Designed for Amazon Alexa

January 11th, 2018 2 comments

Allwinner unveiled their SoC-Only 3-Mic Far-Field Development Kit for Amazon AVS last week, but they are now joined by another low cost silicon vendor as Amlogic has just launched their own A113X1 far-field dev kit officially support for Amazon Alexa Voice Service (AVS).

The development kit is powered by Amlogic A113X SoC designed for such applications with “an audio pipeline that supports high fidelity audio with soft DSP algorithms for both frontend and backend processing”.

 

Amlogic A113X1 far-field devkit specifications:

  • Mainboard
    • SoC – Amlogic A113X quad core Cortex A53 processor
    • System Memory – 512MB DDR3
    • Storage – 512 MB NAND flash
    • Connectivity – SDIO WiFi/BT (AP6356S)
    • Audio
      • SPDIF_IN jack
      • LINE_IN/LINE_OUT jacks
      • 2x Audio headers (MIC_Connector & SPK_Connector)
    • USB – 1x micro USB 2.0 OTG port
    • Expansion – SPI header
    • Misc – 6x ADC Keys, IR_IN/IR_OUT, UART Interface (RS232), LEDs
    • Power Supply – 12V/2A
  • Microphone board
    • 6x digital microphones in a circular array
    • Texas Instruments PCA9557PWR IO expander
  • Speaker board
    • Texas Instruments TAS5707PHPR 20-W Open-Loop Stereo Digital Input Class-D Audio Amplifier with Speaker EQ and DRC
    • Power Supply – 12V DC barrel jack

The solution is said to run “high-performance DSP algorithms for acoustic echo cancellation, beamforming, and noise reduction”.

 

Beside the three boards of the kit (main, speaker, and microphone), you’ll also get a power supply, a serial debug adapter, and a pair of generic speakers. You’ll find more documentation, a getting started guide (with a Linux 4.9 buildroot based distribution), and a purchase link for the $250 kit on a dedicated Amazon Developer page. The kit is currently demonstrated at the Amlogic suite in the Venetian (Suite #34311) during CES 2018.

We’ll also find the kit in company of the aforementioned $129 Allwinner Amazon AVS kit, a new $1,250 “Qualcomm Smart Audio 6-Mic Development Kit for Amazon AVS”, and as well as the $299 hands-free “Synaptics AudioSmart 2-Mic Dev Kit for Amazon AVS” on the System Dev Kits section of Amazon AVS Development Kits page.

 

Amlogic Far-field Kit Accessories – Click to Enlarge

 

Thanks to Theguyuk for the tip

NVIDIA Xavier AI SoC Now Sampling, DRIVE IX & DRIVE AR SDKs Announced

January 8th, 2018 1 comment

Well over a year ago, NVIDIA introduced Xavier, their next generation self-driving and artificial intelligence processor, with eight custom ARM cores, a 512-core Volta GPU, and support for 8K video encoding and decode. A few months ago, the company provided some more details and unveiled NVIDIA DRIVE PX Pegasus A.I. computer for level 5 autonomous driving with two Xavier processors and two NVIDIA next-generation GPUs delivering a total 320 TOPS of computing power. For that it’s worth, 320 TOPS is about 3200 times more powerful than Intel Movidus Neural Network Compute Stick.

CES 2018 has now started, and NVIDIA made several announcement related to gaming and automotive markets, and confirmed Xavier is now sampling to select customers.

Click to Enlarge

What’s really new from the announcement is the addition of two new SDKs (software development kits) for the processor beside the original NVIDIA DRIVE AV autonomous vehicle platform:

  • DRIVE IX – Intelligent experience software development kit that will enable AI assistants for both drivers and passengers using sensors inside and outside the car.
  • DRIVE AR – Augmented Reality SDK designed for interfaces that deliver information points of interest along a drive, create alerts and navigate safely and easily.

This type of powerful hardware and software is however reserved to automotive customers, so most of us won’t be able to get hold of such platform, but we may end up being users of the technology soon enough, as NVIDIA announced partnerships with Volkswagen, Uber, ZF tier-one automotive supplier working with Baidu, and Aurora, a US startup designing and building self-driving technology.

Samsung 8K QLED TV Uses Artificial Intelligence to Upscale SD Content to 8K Resolution

January 8th, 2018 4 comments

You’ve probably read how artificial intelligence can be applying to modify photos, either repair them, or change the weather in photos from sunny to rainy, and vice versa. Samsung unveiled an 85″ 8K QLED TV right before CES 2018, which should be news by itself, but that I found really intriguing is the TV’s artificial intelligence capabilities that will upscale standard and other resolution content to 8K resolution.

The TV will take SD, Full HD or 4K sources runs it through machine learning algorithm to re-create details, reduce noise, and restore edges to bring them back to 8K picture quality. Not your usual “dumb” upscaling 🙂

The TV is said to be equipped with a database that studies and analyzes millions of images in advance to transform low-resolution content into high-resolution, and capable of selecting the optimal filter to convert the source to higher quality. Samsung then explains the images is processed 64 times (how?) to offer natural images in high resolution, and finally elements of picture quality (black/blooming/brightness) are categorized in the input sources to produce pictures with more detailed contrast.

Samsung’s AI technology can also optimize audio for specific scenes, for example the audio from spectators may be amplified while watching sports events, or low-frequency sounds be highlighted while watching concerts.

This will be interesting to find out how well that works, and if it does we can eventually (i.e. in several years) expect some changes in the way video is streamed. For example, if somehow it becomes possible to obtain 8K quality while streaming SD quality videos, then content providers would then likely send low bitrate SD resolution data to compatible devices, and let them upscale content to 8K or whatever native resolution the TV supports. That seems to good to be true so wait and see. Alternatively (more likely?), this could become a fallback option when network connectivity is not good enough for 8K content.

While Samsung is showcasing an 85″ 8K TV at their event, the company will also launch 65″ and greater 8K QLED “A.I.” TVs in H2 2018. The company also introduced that they call “the wall” at the event, a 146″ modular TV featuring self-emitting MicroLED technology promise lower power consumption, and longer life time.

Rockchip RK3399Pro SoC Integrates a 2.4 TOPS Neural Network Processing Unit for Artificial Intelligence Applications

January 8th, 2018 7 comments

Rockchip RK3399 (aka OP1) SoC was launched in 2016 with an hexa core Arm Cortex A72/A53 processor, Mali-T860MP4 GPU, support for 4K video decoding, and high speed interfaces like USB 3.0 and PCIe, as well as Gigabit Ethernet. The processor is found in Chromebooks, TV boxes, development boards, and other devices.

The company has unveiled an upgraded “Pro” version of the processor at CES 2018. Rockchip RK3399Pro appears to have most of the same features as its predecessor but adds a neural network processing unit (NPU) delivering up to 2.4 TOPS for artificial intelligence and deep learning applications.

The company claims that compared to traditional solution, the computing performance of typical deep neural network Inception V3, ResNet34 and VGG16 models on RK3399Pro is improved by almost one hundred times, and power consumption is less than 1% than A.I. solutions implemented using GPU acceleration.

Based on the information provided in the chart above (source: elDEE on twitter), Rockchip RK3399Pro outperforms other high-end SoCs (for such tasks) including Apple A11, and Huawei Kirin 970, both of which also features an NPU, and even offers better performance than NVIDIA TX2.

RK3399Pro NPU supports 8-bit & 16-bit operations, is compatible with various AI software frameworks and APIs, including OpenVX and TensorFlow Lite/AndroidNN API, and comes with AI software tools capable of handling Caffe/TensorFlow models. An RK3399Pro hardware reference design can also be provided to speed up development, but no details were provided.

Click to Enlarge

 

Via Liliputing

$129 Allwinner R18 based 3-Mic Far Field Amazon AVS Development Kit in the Works

January 4th, 2018 17 comments

Several companies are already offering development kits for Amazon AVS (Alexa Voice Service), but as we’ve seen in the past, those are rather expensive with far-field kits such starting at $349 with kits such as Synaptics AudioSmart 4-Mic Development Kit, or Intel Speech Enabling Developer Kit, and hands-free kits being barely cheaper at $299 and up.

But there will soon be a cheaper solution, as Allwinner and SinoVoIP (aka Banana Pi) are working on “SoC-Only 3-Mic Far-Field Dev Kit for Amazon AVS” that includes 3 microphones, and works without special DSP, relying instead on Allwinner R18 processor’s audio codec and capabilities.

Click to Enlarge

Allwinner SoC-Only 3-Mic Far-Field Dev Kit for Amazon AVS (aka R18-AVS-EVK) specifications:

  • SoC – Allwinner R18 quad-core Cortex-A53 processor @ 1.15GHz with Mali400MP2 GPU
  • System Memory – 1GB DDR3
  • Storage – 8GB eMMC flash
  • Video Output – HDMI
  • Audio – 6x Microphones, 2x AEC, AUX and headphone output; GMEMS voice recognition algorithm
  • Connectivity – Dual band WiFi, Bluetooth 4.0
  • USB – 1x USB type A port, 1x micro USB OTG port
  • Power Supply – 12V DC input
  • Dimensions – Mainboard: 100 x 100mm; microphone array board: 90 mm ∅

The board will support Linux operating systems at first, but Android is also being worked on. A ribbon cable is also included in the kit to connect the mic array to the main board. Now you may wonder why a 3-mic development kit comes with 6 microphones. Allwinner explains:

6 microphones are included on the board, while only three are used and qualified, providing flexibility to tune for 6/4/3/2 mic solutions and freely match with different product designs

I’m not 100% sure what that means, but I guess the kit only works for three for now, but in the future algorithms may support a combination of up to 6 microphones. We’ll have to see how the solution works compared to DSP based systems.

Allwinner R18 Block Diagram

The development kit is now found on Amazon website yet, but a page on Banana Pi website mentions they are 50 unit for pre-sale for $129 with shipping scheduled on February 5, 2018. The “Buy” link does not work yet. A few more details may be found in the product page on Allwinner website.

Samsung Announces Exynos 9810 Octa-core Processor Optimized for AI and Multimedia Applications

January 4th, 2018 3 comments

Samsung Electronics has just announced the launch of Exynos 9 Series 9810 (Exynos 9810) manufactured with Samsung’s 10-nm FinFET process, featuring an eight core processor clocked up to 2.9 GHz, a gigabit (1.2 Gbps) LTE modem and deep learning-enhanced image processing.

Single core performance is aud to be improved by two-fold, while multi-core performance gets a 40% improvement compared to the previous generation chip, which should be Exynos 8895. ARM Mali-G72 GPU is said to bring more realistic graphics along with 20% more performance.

Samsung Exynos 9810 specifications with extra cache and memory info from Anandtech:

  • CPU
    • Quad core custom Exynos M3 @ up to 2.9GHz optimized for performance; 512KB L2 cache per core
    • Quad-core Arm Cortex-A55 @ up to 1.9GHz optimized for efficiency; 128KB L2 cache per core
  • GPU – Arm Mali-G72MP18
  • Memory – LPDDR4x (4x 16-bit @ 1794 MHz)
  • Storage – UFS 2.1, SD 3.0
  • Display –  Up to WQUXGA (3840×2400), 4K UHD (4096×2160)
  • LTE Modem – LTE Cat.18 6CA 1.2Gbps (DL) / Cat.18 2CA 200Mbps (UL)
  • GNSS – GPS, GLONASS, BeiDou
  • Camera – Rear 24MP, Front 24MP, Dual Camera 16+16MP
  • Video – 4K UHD 120fps recording and playback with 10-bit HEVC (H.265), H.264, VP9 Codec
  • Process – 2nd gen. Samsung 10nm FinFET Process

The company did not provide much details about deep-learning acceleration, except it will leverage hardware and software…:

Exynos 9810 introduces sophisticated features to enhance user experiences with neural network-based deep learning and stronger security on the most advanced mobile devices. This cutting-edge technology allows the processor to accurately recognize people or items in photos for fast image searching or categorization, or through depth sensing, scan a user’s face in 3D for hybrid face detection. By utilizing both hardware and software, hybrid face detection enables realistic face-tracking filters as well as stronger security when unlocking a device with one’s face. For added security, the processor has a separate security processing unit to safeguard vital personal data such as facial, iris and fingerprint information.

The Exynos 9 Series 9810 is currently in mass production, and should be found in smartphones, personal computing devices, and automotive products later this year. More details can be found on the product page.

Laceli AI Compute Stick is a More Powerful & Efficient Alternative to Intel/Movidius Neural Compute Stick

January 3rd, 2018 3 comments

Intel’s Movidius Neural Compute Stick is a low power deep learning inference kit and “self-contained” artificial intelligence (A.I.) accelerator that connects to the USB port of computers or development boards like Raspberry Pi 3, delivering three times more performance than a solution accelerated with VideoCore IV GPU.

So far it was the only A.I USB stick solution that I heard of, but Gyrfalcon Technology , a US startup funded at the beginning of last year, has developed its own “artificial intelligence processor” with Lightspeeur 2801S, as well as a neural USB compute stick featuring the solution: Laceli AI Compute Stick.

The company claims Laceli AI Compute Stick runs at 2.8 TOPS (Trillion operation per second) performance within 0.3 Watt of power, which is 90 times more efficient than the Movidius USB Stick that can deliver 100 GFLOPS (0.1 TOPS) within 1 Watt of power.

Information about the processor and stick is rather limited, but Gyrfalcon explains their APiM architecture (AI Processing in Memory) uses memory as the AI processing unit, which eliminates the large data movement resulting in high power consumption. The processor comes with 28,000 parallel computing cores and does not require external memory. The stick is also equipped with 4GB storage, a USB 3.0 port, and works with Caffe and TensorFlow.

Movidius Neural Compute Stick vs Laceli AI Compute Stick

Going to the company website, we’ll also find a complete development kit with USB Interface, eMMC flash, and special access port, as well as a multi-chip board with PCIe and M.2 Interfaces that appears to combine eight Lightspeeur 2801S AI processors.

The processor is already in production, and available to “qualified customers”, while the Laceli AI Compute Stick will first be showcased at CES 2018 in Las Vegas in a few days.

Thanks to TLS for the tip.

Google Assistant SDK Now Supports Device Actions, More Languages (French, German, Japanese)

December 29th, 2017 1 comment

Back in May 2017, Google released the Assistant SDK that worked on Raspberry Pi 3, and other ARM boards, essentially transforming low cost development boards into Google Home equivalent. The SDK became more popular once Google’s AIY Voice Kit was launched since it offered an easy and inexpensive way to use it with Raspberry Pi 3 board.

Since all you need was a Linux board with an Internet connection, a microphone, and speaker, I tried Google Assistant SDK on one of the cheapest platform available: Orange Pi Zero Set 6 Kit including Orange Pi Zero board, but also an expansion board with built-in microphone and audio output jack, and a cute little case. I added my own pair of speakers, micro SD card, and USB power supply, and after setting up the software, I was able ask question, and get answers with female voice using the demo app.

At the time however, there was some limitations, as integration with home automation devices was not easy, English US was the only language option, and we were stuck with a female voice. Since then, Google has added support for male voice for text-to-speech, and as I checked the release notes, Google added support for more languages, and device actions in December 20.

Changelog:

  • Google Assistant Library (developer preview 0.1.0)
    • Support for Device actions.
    • Support for more languages: English (Australia, Canada, UK, US), French (Canada, France), German, and Japanese. Selectable from Google Assistant app.
    • Location can now be configured as a street address in the Google Assistant app.
    • Better handling of connection errors.
  • Google Assistant Service (v1alpha2)
    • Support for Device actions.
    • Support for more languages: English (Australia, Canada, UK, US), French (Canada, France), German, and Japanese. This setting can be passed through the Service API or selected from the Google Assistant app.
    • Location can now be configured as a street address in the Google Assistant app, or as a latitude and longitude via the API.
    • Support for displaying the text of the user’s request and the text response from the Google Assistant.
    • Support for submitting queries via text input (Using Device actions or IFTTT).

I could update the library on Orange Pi Zero as follows:

I could still use my DIY smart speaker to ask questions and get answers after a reboot, so the update went smoothly. Controlling other devices like Sonoff TH16 will require some more studying.