Feature Article:

Speech and warehouse management
Talk to me

By , independent journalist

Speech recognition systems allow for natural language data collection and hands-and-eyes-free operations. In a multitude of applications, such as warehouse management, it allows to perform various operations simultaneously, for example data gathering and pick and place. If well designed and integrated into other processes, speech input can speed up operations significantly, increase accuracy and improve comfort and safety in the workplace. The following article discusses how ?speech? can contribute to more productive logistics and warehousing systems, an absolute requirement of the 24/7/365 e-conomy.

In 1985 Westinghouse built a speech recognition terminal but it took another two years before a portable ?talkman? was installed at Ford MC in the USA. The first applications were in the area of quality control. Car manufacturer FMC used speech in its painting control process and weapons builder Raytheon in the microscopic inspections of printed circuits.

At first systems were quite slow and required clear articulation. By 1991 the machine had taken up a ?normal? speaking pace and by 1994 the ?talkies? had been coupled to narrow band radio. Now operators could ?talk on the walk?.

Ever shrinking delivery timeframes ?e.g. a delivery on Saturday instead of on a Monday, for an order placed on Friday? builds on ever faster processing in warehousing

The technology hit Europe in 1994, particularly in warehousing applications. The use of spread spectrum allowed for very compact RF-equipment (radio frequency) that easily integrated with warehouse management and logistics systems. The technology is now taking up speed in this particular area where competition is fierce and margins are small, where operations are labour intensive and tasks repetitive. Further shrinking of delivery timeframes ?e.g. delivery on Saturday instead of on Monday for an order placed on Friday, or 24-hour delivery for goods purchased through the Internet? builds on ever faster processing in the warehouses. Speech technology can contribute to that goal.

In terms of hardware, speech systems comprise a combination of microphone and headset, a processing module and an RF-transmitter/receiver that links to the server and databases. In the future operators may also wear visual display units ? goggles ? that enhance ?reality? for more complex, procedural and even learning operations. Now the information comes from the warehouse management system (WMS) and ERP-system in the form of ?spoken? questions and instructions. Mobile workers give vocal feedback to this interactive system.

Background noise

Performance degradation of voice recognition systems in noisy, real-world environments - such as warehouses - is probably one of the most significant factors limiting take-up of the technology today. This plays an important role in the accuracy of the recognition system and has a direct bearing on the acceptance of the technology by end users.

Solutions to noisy operational environments come from hardware improvements - directional, noise filtering microphones, and also from improved speech recognition algorithms. One example of a current EU funded research project taking the latter approach is project RESPITE, whose aim is to develop novel techniques for speech recognition that are robust to ambient noise and signal deterioration.

Another approach to improving speech recognition accuracy comes from two different perspectives of the speech signal - speaker dependent and the speaker independent. In the speaker dependent approach the system first learns to recognise an appropriate set of utterances by the individual speaker. On the other hand, speaker independent systems are ?language dependent?. They recognise ?language? rather than different voices.

The speaker dependent approach uses relatively less expensive technology but requires more training time. It takes the system some ten minutes to ?register? the voice profile of a user. It is independent of language or dialect... as long as ?sound? and ?meaning? have been matched beforehand. When the operator has a cold the profile needs to be adapted.

The speaker independent system is open to a larger variety of voices within a given language set, without the need for training or re-training. That could be an advantage in case of frequent personnel moves, e.g. temps. And there is no adapting necessary in case of a cold.

It would appear that the fast growing market for speech systems focusses less on the particulars of technology but on their seamless integration in the logistic process. Integrators now choose the most suited technology for the job. Return on investment (ROI) has become an overriding criterion.

?

Speech at work - Reed Boardall

Case 1 - Reed Boardall: ?We surely expect a lot?

"We don't work in figures," says Gary Tilburn, managing director of Reed Boardall, one of UK's biggest logistical service providers who handles over 1,5 million pallets a year for a turnover of more than ? 1 billion. As volumes vary daily, it is difficult to compute the exact benefit of the new technology. "Since we introduced speech technology in the cold stores, in February 2001, we need fewer pickers. About one third less. That means one third more work is done and our pickers spend less time in the cold." The company started with ten picking terminals. Nowadays, at any one time, three to four terminals are not in use. "That's how efficient the system is." Procedures have not changed. But there is no need to use pen and paper any more in the -28?C stores. "Change is dangerous." Pickers were involved from the start and formulated how they wanted to run through the speech sequences. Implementation proved to be anything but complicated. A methodology was agreed upon with Chess Logistic Technologies, the provider of the Emperica WMS-system, who integrated the SyVox ISIS-interface in its program. A link was established within a couple of days. One warehouse after another was brought online. "We could have gone back to pen and paper any time." Nothing went wrong. In fact Reed Boardell recently signed a new contract for forty extra sets. The speech technology will also be introduced further up- and downstream in the process chain, for loading and unloading trucks. Reach-trucks drop their pallets in the central corridor. Pedestrian vehicles then pull one or two pallets to the thirty loading docks. Some warehouses are 150 meters long. In the meantime the operator reads out the pallet numbers and the system responds with their exact destination. The system correlates pallet, loading bay, loading number and trailer number. "We do not yet know how efficient this new application will be, but we surely expect a lot."

?

When to go for voice?

When to choose voice rather than opt for other data collection systems, such as bar code scanning? Voice may be quicker, more accurate, convenient and safer when the operator is required to manipulate products, e.g. in order picking, and additional or variable information is required or must be passed on to the system (such as gate number, corridor, column, number, weight, pallet...).

Recent implementations in ?cold stores? make life much more comfortable for operators. They no longer need to use paper and pencil. Voice can be complemented with other data gathering technologies (barcode scanning, RF chip...) when very complicated data, such as long series of numbers are to be transferred. ?Speaker dependent systems are very suitable for input of more complex series of numbers and catchweight-picking,? says Rodolphe Becker, director of VoCognition.

?

Case Study - alli

Case 2 - alli: Absolutely flawlessly

"With SpeechNet Logistics by Syvox we've reduced errors from 3% to 0.3%," says Otto Pschera, project leader Automation & EDI Technology at alli. The quality benefit is staggering when one takes account of the volume: 10.000 units a day. "A true revolution." alli evolved from being a logistics player in the fresh food sector to a fully-fledged supplier in the value chain. The company organises some 2000 deliveries daily to over 5200 points in Eastern and Northern Germany. Since 1999 it also delivers freshly packed meat to retail outlets. Just in time. The throughput also increased with 50 to 100 percent, especially in the variable weight/pricing area. "Before we had to deploy two people. One person would take hold of the product, weigh it and call out the results. The other would write them down and afterwards the same figures would be typed into the system. Now one person reads the weight directly into the system. Correct to the gram." All this results in a very speedy return on investment. "Three months," says Pschera. If it were not for the extraordinary combination of speech and the weighing process ROI would be estimated at 12 to 15 months. The introduction of speech technology lead to better performance, higher quality, easier maintenance and higher worker satisfaction. "Clearly errors result in problems. People who make fewer errors, perform better. They are praised instead of being reprimanded. They are much more satisfied now that they make fewer errors." The equipment is easy to wear and the ergonomics of the job has improved. "If we would take out the system, I am sure it would result in protests on the floor." In principle the speech system is easily implemented. "We experienced a couple of hiccups," Pschera recalls. When alli brought the system online at the end of 2000, it was the first speech system in the German language. Each language has its particularities that speech recognition systems have to integrate. It took some time to optimise the application. There was also a problem with the microphones that the people wear. These had been tested and performed well in temperatures above 6?C and even at minus 28?C. "But in areas roundabout the freezing point, i.e. between plus and minus 1?C, we experienced problems." Water vapour of one's respiration would freeze onto the equipment and disrupt the sound. A repackaging of the mikes solved the problem. "We were also the first to go online with a highly dynamic expedition service." The speech solution had to be integrated in the existing WHM-system and in the aKS picking system already deployed. There was a learning curve, though. The system reached its full efficiency in the following three months and has been running absolutely flawlessly since September 2000. "We are now developing concepts for the introduction of speech support in other areas," says Pschera. In other expedition areas and in the logistics process, e.g. incoming goods, warehouse reorganisation and stock monitoring "where we mainly expect quality improvements."

?

Speech market

Speech is now rapidly moving out of the ?early adopters? into mainstream markets. Early adopters were probably ?technology huggers? in SME-operations. The retailer chain Mestdagh in Gosselies (Belgium) runs the VoCognition system in its 8.000 m2 fresh produce warehouse since 1997. ?I was their first customer,? smiles Michel Eggermont, EDP manager who equipped a dozen pickers. The advantage? ?We move a third more.? Belgian wine merchant Le Clos du Renard was the first to implement the French-language SyVox product. Analysts of Deloitte & Touch calculated there that the picking error rate was reduced from 0,007 to 0,000033, validated complaints fell from 70 to 1 per week and picks per hour increased with 30%.

?The number of speech users quadruples each year,? says Jan Vermeesch, European marketing manager of SyVox. "SyVox [currently] supports English, Spanish, French, Dutch, Flemish, German and Italian."

?The number of speech users quadruples each year,? says Jan Vermeesch, marketing manager Europe at SyVox (Brugge, Belgium), manufacturer of interactive speech applications for mobile workers. "SyVox opts for a speaker independent technology and now supports various languages: English (US and UK), Spanish (Mexico and Spain), French, Dutch, Flemish, German and Italian."

Becker at VoCognition agrees: ?Voice is hot!? The company now has some thirty installations running in Europe. ?Last year our turnover increased fivefold.? In the current lifecycle however technology remains the door opener. ?Later we can show we?re capable of implementing a whole system. Ultimately we talk about ROI,? says Becker who expects the technology to jumpstart soon. ?Huge operations have decided the technology works and they will go for it with hundreds of terminals at a time. Voice is a strategic decision. These customers will deploy the technology across the continent. Smaller partners and customers will follow suit.?

?Acceptance barriers are lifted through standard interfaces that integrate with industry standards in ERP and WMS,? says Vermeesch at Syvox. Implementations are rapid, near plug & play. ?The logistics market has not slowed its investment decisions after September 11th. But the sector invests only if ROI is less than 12 months,? says Vermeesch. And that is exactly what speech systems are capable of.

The better ROI is achieved in order picking in difficult environments. New applications, such as replenishment, supply, cross docking and reception, are already available or in development. The technology has also been used in quality control in a conveyer or assembly line environment. Integrators now follow their customers abroad.

?

Unilever - Iglo-Ola

Case 3 - Iglo-Ola: Increase picking quality with voice

?Our aim was to increase the quality of our picking operation. We?ve succeeded in reducing picking errors by 70%,? says Willy Baeyens, Cold Store Operations Manager at Iglo-Ola, a major ice cream and frozen food distributor in Brussels and part of the Unilever Group. Some 30 pickers, who operate in a -25?C environment, have now been equipped with VoCognition sets. The benefit in speed was less important, some 5%. ?But then again we already had optimised the process and quality was the focus here, not speed.? That also impacts on the ROI, which is calculated to be two years. The financial profit comes from savings in re-picking, extra deliveries, administration,... ?We are certain our customers are more satisfied, too.? The system chosen - VoCognition - is ?speaker dependent?. That means the user must ?train? the system which stores his/her voice template on the server. This ?one time? procedure might take about a quarter of an hour, followed by some fine tuning afterwards. ?That also allows first time users to familiarise themselves with the system.? This technology is reported to be more accurate than ?speaker independent? systems when used by people with speech-defects. During the testing phase some 40% of the pickers had used the system and shared their experiences and ideas, contributing to the final solution. ?This early involvement in the process undoubtedly increased the acceptance by the users,? says Baeyens whose ultimate goal is ?errorless picking?. In the future the system will be linked with SAP, the WMS and integrated to the customer service level monitoring. Baeyens is delighted with the results to date, but does not consider voice picking to be the ideal solution for just everything. ?We?ll prefer to manage our pallets with traditional terminal systems on our forklifts.?

?

What's in it for me?

It is the task of integrators to optimise the process and the technology. ?There can be no box-moving here,? says Vermeesch. Each implementation requires services to harmonise the dialogues and the operations. For example, the system should not repeat the full co-ordinates of the location if it directs the picker to exactly the same location. ??Same location? will do fine.?

Communication, change management and training require the necessary attention if the system is to deliver. Product marketing manager Bart Rivi?re at Syvox warns that the technology is an enabler. Pick-it champions therefore will generally be enrolled in project teams. ?Ultimately the people who use the apparatus make the difference. Make sure to clarify what?s in it for them.?

?

? ? ?

Luc De Smet

is an independent journalist, with a background in peace research, having worked for Pax Christi Flanders, for International Peace Information Service in Antwerp, and later at the Life & Peace Institute in Uppsala (Sweden). He writes on a broad range of topics including the economy, management, human resources, industry, logistics, environment and technology matters.

The editors of HLTCentral would welcome any feedback on the article.
Please send your comments to the .

Disclaimer: Any opinions expressed in this article are strictly those of the author and do not necessarily reflect the viewpoint of HLTCentral or its editors. Copyright ? 2002 HLTCentral. All rights reserved.