Suppliers make voice-directed work even more productive with analytics and optimization, and improvements to voice recognition engines.
Suppliers of voice-directed systems are quick to point out how voice fits well with today’s e-commerce picking and fulfillment challenges because of the “hands-free” nature of voice directives versus using paper pick lists or wireless handhelds. By using Bluetooth headsets to hear verbal instructions and validate information back verbally to the system, voice keeps workers’ hands free for work tasks.
But voice productivity doesn’t stop there. In recent years, vendors have been bumping up the productivity of their solutions by doing things like applying analytics and optimization software to find further efficiencies, making improvements on speech recognition, and leveraging new device capabilities.
Vendors are also working to make voice recognition more effective. While tech giants such as Apple (with Siri) and Amazon (with Alexa) are making voice a growing part of the consumer experience, voice-directed solution vendors for the warehouse say that these consumer-focused voice engines aren’t yet suitable for warehouse environments.
With the recent release of its Lydia Voice 8 voice system, Ehrhardt + Partner Group (EPG) has added neural network technology to improve voice recognition performance, according to Scott Deutsch, E+P’s North American president. It’s similar to the neural network technology used by Amazon for Alexa or by Apple with Siri, but tuned for the rigors of noisy warehouse environments, Deutsch says.
“With consumer voice technologies, most of you know that you sometimes end up having to repeat a question, but that’s just unacceptable in a warehouse environment where near-perfect voice recognition is required,” Deutsch says. “Warehouse workers simply can’t waste time repeating things.”
However, adds Deutsch, E+P developers were able to use neural network technology—which uses high computing power to analyze large data sets to look for patterns—and adapt it to supply chain needs. E+P used “pruning” algorithms as part of neural network development to narrow down large data sets to find ways to improve voice recognition response time as well as to quickly understand tricky dialects, says Deutsch.
To improve Lydia’s understanding of mixed dialects, says Deutsch, E+P developers applied neural network technology against data from a customer site in Switzerland that had a tricky mixed dialect of German and Swiss inflections in workers’ voices. The new version offers about a 25% performance improvement in voice recognition over previous generations of the system, says Deutsch.
For workers, improved performance means quick interactions with the system with fewer instances of having to repeat information, says Deutsch. The improved engine also eliminates the need to use voice template training with new users, while also cutting the need to retrain the system should a worker’s voice change due to an issue like a sore throat.
Another recent enhancement to E+P’s Lydia is a function called Enterprise SiteSwitcher, which enables an IT department to set up and manage voice system security and profiles centrally. The devices used with the system are smart enough to adapt to each location’s security details, which allows for central device configuration with the ability to easily move devices from one location to another.
While the e-commerce picking pressures are making voice solutions attractive, vendors need to respond by making their systems easy to deploy, adds Deutsch. “Because of e-commerce’s peaks and valleys, and the struggles operations are having in finding people, voice solutions need to be extremely easy to roll out and get new users operational with rapidly,” he says.