Children are on Voice-Enabled Devices. Edge Voice-AI is the Solution to Privacy Risks.

Kadho Inc
Kadho Inc
Sep 8, 2018 · 6 min read

Human brain processes information on the “Edge” and not “Cloud”. There are benefits to performing AI processes such as Voice-AI on edge including: privacy, latency, reliability and of course cost!

Industrial companies, for example, are developing AI, machine learning and other technological innovations to empower new levels of productivity and efficient performance. While cloud computing has served as a major enabler of this transformation, edge computing is fast becoming a key component of the next generation of this digital transformation.

AI engines’ web APIs are able to respond in sub-seconds via broadband internet connectivity. This works quite well for prescriptive and predictive analytics, including sales forecasts, product suggestions, composing music, disease diagnostics, processing job application data and much more. For real-time operations, however, where every millisecond counts, this not fast enough, especially where nothing less than real-time response is critical. Furthermore, sending data to the cloud creates privacy and security issues.

For example: a robot surgeon performing sensitive operations on patients that needs to examine images and make virtually instant decisions every second; driverless cars that have to navigate around obstacles and move through complex situations and terrain; police surveillance cameras that need to analyze and process images. In countless situations in every industry, massive amounts of information is being sent to the cloud that may hold PII (personally identifiable information), breach privacy laws and be vulnerable to theft.

Alternatives such as remote locations and proprietary data centers all have their limitations especially as data from IoT devices increases exponentially. Cisco Systems, for example, forecasts that cloud traffic will rise nearly fourfold in the next few years, increasing from 3.9 zettabytes (ZB) per year in 2015 to 14.1ZB per year by 2020. IoT data growth itself could cause the perfect cloud computing storm.

The Role of Edge Computing

Edge computing is data processing that takes place at the edge of the network instead of in a cloud or a central data warehouse. Edge computing enables knowledge generation and analytics to happen at the data source. This means that devices may not be connected to a network continuously. To date, the primary role of edge computing has been to process, filter, store and send data to the cloud. With edge AI, algorithms are executed locally on the hardware device using sensor signals or data is created on the device. In addition to those already uses already mentioned, Edge AI is advantageous for applications such as smart traffic lights, ATMs to stop fraudulent financial transactions, retail stores for in-store incentives and others.

Edge computing will not replace cloud computing. Analytic models or rules may be created in a cloud and then pushed to edge devices. Using a combination of cloud and edge computing, big data, machine learning and advanced analytics in operations, industrial companies can improve asset performance, lessen unplanned downtime, lower maintenance cost and create business models that could potentially capture new value from machine data. Industrial companies have recently started incorporating both cloud and edge computing into operations to collect insights from huge data volumes that assist in achieving key business outcomes, for example, reducing unplanned downtime and lowering energy consumption.

As more computing, storage, and analytic capabilities are pushed into smaller devices that reside closer to data sources, edge computing will be crucial to enable edge processing to deliver on the promises of the IoT.

A number of key drivers in the shift to edge computing include:

  • Increased computing power available in devices with a small footprint, e.g. gateways or sensor hubs.
  • Computing and sensor prices continue to plunge.
  • Modern analytics and machine learning improvements advance.
  • Increasing data volumes generated by machines, e.g. market pricing or weather data.

Industrial companies increasingly will find edge computing crucial in the following areas:

  • Low or intermittent connectivity
  • Low latency, including closed-loop interactions between machine perception and taking action.
  • Bandwidth and the high costs associated with transferring data to the cloud.
  • Access to time-based data used for real-time analysis.
  • Immediacy of analysis.
  • Regulatory, cyber security and compliance constraints.

Business implications of edge computing technology already are convincing. The Edge Computing Consortium recognizes the following potential outcomes:

  1. Predictive maintenance : Reduced costs, Assurance of security, Extension of products to service
  2. Management of Energy Efficiency: Reduced energy consumption, Reduced maintenance costs, Increased reliability
  3. Smart manufacturing: Reduced product life due to increased customer demands, Production mode customization, High-volume manufacturing replaced by multi-batch and small-quantity modes
  4. Flexible replacement of devices: Rapid new process and model deployment, Flexible production plan adjustments

Edge Computing and Speech Recognition

Wearable devices and smartphones nowadays offer speech recognition as a standard feature for search, dictation and voice commands. The majority of studies on improving accuracy for these functions assume that speech recognition will be done on powerful servers in data centers. Despite increased availability and speed of the mobile internet, speech recognition tasks still have high latency or even fail completely due to network connections that are unavailable or unreliable. A speech recognition system that is embedded and runs on mobile devices will have lower latency and be more reliable. It must, however, be accurate while not consuming significant computational resources or memory.

Edge Voice AI

Edge Voice AI solutions such as KidSense.ai do not send speech data to the cloud, but perform closer to where the speech is originates. Currently, state of the art speech recognition systems use huge computing resources and need to be on the cloud. Bringing speech recognition to the edge is a challenge due to edge devices having limited computational resources. To achieve high accuracy and low latency at the device level requires inventive transformation of the current architecture of speech recognition. Deep neural networks thrive on big data, but as a result these networks and their requirements need to be bigger. Edge Voice AI needs to create networks that are more efficient and streamlined, and use less computational power and data. To attain Edge Voice AI with neural networks, these networks need to be quantized and provide small language models that can be changed dynamically.

Edge Voice AI Advantages

When edge employs processing where the user exists and data is captured, this reduces the need for centralized remote computers and has several unique advantages:

  • Decreased operating cost due to less data management and operational needs.
  • Real-time data analysis as data is analyzed at the device level and less dependence on core cloud computing.
  • Improved application performance as applications interact directly where the data is collected and processed.
  • Zero or low network traffic since no data is transferred to a cloud, thereby reducing data transfer bottlenecks leading to zero latency.

Increased privacy of data and improved latency benefit users as Edge Voice AI reduces cloud and data transfer dependencies. Privacy is a major concern in this cloud era as illustrated by numerous publicized cases of invasion of privacy:

  • Facebook Cambridge analytics.
  • Mattel pulling Aristotle from the shelves.
  • Concerns over Alexa eavesdropping.
  • Vtech being fined for violated COPPA.
  • Privacy concerns over Google home.
  • Europe’s stricter legislation for data protection.

These privacy issues are eliminated when data does not leave the device while, at the same time, latency will be improved dramatically as data is processed on the device. Edge Voice AI can provide privacy protections in a child’s room where technology provides entertainment, helps a child to develop, and provides peace of mind for parents. If the room contains, for example, a robot with a camera as well “smart” furniture, parents should be concerned about compromises of the child’s privacy if these devices are connected to the cloud where hackers can access cloud storage. In addition, Edge Voice AI becomes important for protecting child’s activities from businesses seeking data for market research. For more information visit www.KidSense.ai

Kadho Inc

Written by

Kadho Inc

Produces #edgevoiceai based on true#artificialintelligene to enable #kids to #communicate with #technology — #KidSenseai #kidsense

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade