March/April 2005

Voice recognition systems have been simplifying data capture and picking tasks for a long time, but at last they're really taking off. Sharon Clancy considers why

Judging by the high profile voice recognition technology has been getting in the market lately, you might think it was the latest development to hit the logistics market. Actually it's been around for at least fifteen years.

So why all the excitement? Well, it's taken a series of advances in computer hardware, user-friendliness, systems integration capability and the underlying technology to bring voice recognition to its current status as "the next big thing" in warehousing and logistics.

Early systems were not particularly user-friendly. Users had to speak slowly and clearly to have a chance of being understood by the computer. To borrow a bit of telematics market jargon, the "man-machine interface" is now vastly improved.

 

"The market has moved to speaker-dependent systems rather than pure speech recognition systems." That's how recent developments are summed up by Greg Tanner, managing director of Vocollect Europe. "Speech-dependent software, where the user has to train the system to recognise a specific voice, is more useful for applications where there is a high degree of repetition, because it helps the system recognise what the user is saying the first time, speeding up the operation."

Training the computer to recognise each individual worker sounds time-consuming, but the vocabulary range required is very limited. Vocollect reckons it takes only 10 to 15 minutes per worker. The cost and complexity of programming a system to recognise all the nuances of accent and language is prohibitive for low-margin applications such as order-picking.

Another advantage is that if an employee develops a cold and his or her voice changes, the systems can be retrained to recognise the new speech patterns. It's done by simple repetition; the machine says a word and the worker repeats it back.

"It's easier to train people while in the warehouse. If you sit them at a computer (not a natural working environment for them) they will talk differently.

Certainly many of the major players in warehousing systems think so, including hardware specialists such as LXE and Psion Teklogix, and WMS and supply-chain systems suppliers such as Chess, Manhattan Associates, Microlise and Swisslog.

For certain operations, voice technology appears to present an irresistible ROI case. Increased productivity, fewer mistakes and improved health and safety are the most oft-stated benefits. Voice-picking is traditionally associated with fast-picking operations with products such as consumer goods and automotive parts. It also has a role in chill stores, where glove-wearing operatives struggle to press buttons on a scanner, and where the scanners themselves are more expensive (since they have to withstand the low temperatures). On some jobs, operatives need two hands for the actual picking, perhaps because the goods are heavy or bulky (hanging garments are a case in point).

There's no doubt that voice can improve picking accuracy, allowing operatives to make extra checks to confirm correct location, number of items to be picked and so on. It eliminates common problems such as clerical errors in printed lists, or a wrong or missing barcode. There's no slowing down in the operation while pickers who can't find the product search adjacent locations - they can simply speak to the computer to let it know there's a problem.

From paper to voice

Voice system vendors say that many companies now adopting voice are switching directly from paper systems (without having adopted some other computer-based system first). Such an approach is not new; it was seen years ago when one of the UK pioneers of voice technology, PCL, was still active in the market. Through a series of acquisitions, that company's early interest in voice recognition has been maintained by Swisslog, the big supply-chain systems vendor.

Clearly its customers have seen the potential; one of the latest, MDL (the distribution arm of publisher Macmillan), is maintaining the option of integrating voice picking later into a major new Swisslog-based warehouse management system.

Its warehouse operation certainly looks an ideal candidate for such an approach. The roll-out involves 260 staff on three sites, where Swisslog is replacing MacMillan's existing warehouse management system with its own WarehouseManager. This is being implemented with RF picking, so the voice potential is already there; and Swisslog says the use of standard proven interfaces to the voice hardware paves the way for an easy future voice deployment.

So why is everyone not using voice? Well, cost is one deterrent. The cost differential between a traditional handheld RF device (a scanner) and the equipment required for using VR (terminal, headset and belt) can typically be anything from 20 to 40 per cent.

Even if the advantages of voice may seem irresistible, it has to be recognised that the big adopters of RF have so far tended to be companies with high-volume pick operations. If you only have a few pickers, it can be harder to justify the extra cost of voice.

The economic case for VR is also tougher for put-away or replenishment operations. "In most warehouses, there are far more picks than replenishments, " explains Jeff Baum, senior vice president of international operations for Manhattan Associates. "Replenishment involves a whole pallet, box or carton. Picking often means taking just one or two items out of that same box. VR can save you 15 seconds a pick. If you are making 100 picks an hour, that's a significant time saving. If you're putting away 30 pallets, there's less benefit."

Costs are coming down, says Baum, but not exponentially. "There are, however, some impressive incremental financial benefits delivered by VR technology when it is used in the right environment. Voice is ideal for repetitive, high-volume, straightforward picking. Voice is not only hands-free, it's eyes-free as well - employers are not constantly refocusing either on the pick bay or the scanner. "

Rf backbone

All voice systems sit on the basic RF backbone in the warehouse. Manhattan is among those WMS and ERP vendors who can integrate voice directly into their systems. Both Vocollect and Voxware can offer application software that manages not only the voice units, but task assignment and other WMS functions as well.

With other systems, there may be a need to add a middleware layer that allows the WM system to talk to the voice terminal. Middleware is the fast route to voice-directed warehouse working, acting as the link between the WMS and the voice terminals.

Handheld terminal manufacturers such as Psion Teklogix are another useful source of information. They are, after all, accustomed to integrating conventional handheld RF operation with WMS and ERP systems.

VoiteQ, for example, is only four years old, but has already established itself as a key player in the UK market for voice installations in the warehouse. It had developed its own warehouse management system for convenience stores, and deployed its first voice-directed WMS back in 1998. Although it still offers a whole WMS and supply-chain system, its main business now is in middleware installations.

VoiteQ's Voiceman middleware in effect "intercepts" files exported from the WMS, converting them for voice-picking. "On standard picking operations for handheld terminals, the WMS sends a flat file either to a printer in the warehouse or to the handheld," explains John Harper, VoiteQ's sales director. Voiceman works exactly as any other database picking assignment, says Harper, but in addition, check digits are incorporated to ensure the picker is in the right place.

At the end of the pick operation, Voiceman returns the flat file back to the WMS, together with any data on exceptions. "Only completed picks are returned to the WMS. If it cannot be completed for any reason, a supervisor needs to determine why."

The WMS can automatically import data from Voiceman, or it can produce reports or key in data manually. "Choosing the non-automatic route is rare, but it can be useful if, for example, the original supplier of your legacy WMS is no longer in business but the customer wants to carry on using the system."

Managers get nervous about voice technology, admits Harper. "Just remember that your tried-and-trusted WMS is still sorting the pick, and exports it as normal. It doesn't even need to know how the file it has exported is being dealt with, so you can switch back to paper or barcode scanners for particular operations or customers. The Voiceman middleware constantly scans its import file to see if there is anything there."

Middleware is best suited to picking operations, says Harper, where mistakes are more likely to occur. "Avoiding short picks or partial order-fills saves a lot of communication between stores and warehouse, checking when missing stock will arrive. The risk of mis-picks is dramatically reduced, which is crucial if you are handling orders for a multitude of branches. There are also savings in data processing costs."

Health and safety

Although health and safety has not figured greatly in the debate about voice technology, the experts agree that it is now a driver for some companies. "There are concerns about repetitive strain injury, and general warehouse safety issues, such as the need for operatives actually to look where they are going, rather than constantly looking at a barcode scanner to receive instructions on the next pick location."

Manhattan Associates prefers direct integration into the WMS or ERP system. Jeff Baum believes relying on middleware could store up future problems. "Middleware adds complexity and time because it batches commands from the WMS," he maintains. "By eliminating it, the WMS has direct contact with the operative. Task prioritisation and deciding the best equipment for that task are critical elements in maximising warehouse efficiency. With direct integration the tasks can be mixed according to type, operative and time."

Middleware does help legacy systems work with voice, Baum admits. "However, you reach the point where there is a mix of technologies, each with its own middleware. You have to log on to the VR system and then to the legacy system. It all adds time."

Baum also advises managers not to forget the costs associated with voice. "You might reduce the mistakes made by 100 pickers by 10 per cent, but have to hire more administration people to manage it."

Psion Teklogix has pioneered voice technology and is one of Europe's leading suppliers. French supermarkets Carrefour and Casino are two of its biggest customers in this field. "The technology is arousing interest because of the results it produces," says UK managing director Duncan Smillie. "Any company working on narrow margins, doing labour-intensive work with frequently-repeated tasks should consider voice. Accuracy levels are over 99 per cent and picking errors can be reduced by between 70 and 80 per cent. Compared with paper-based systems training times can be halved - useful in a sector with high staff turnover levels. Payback time is usually twelve months."

Psion uses Microsoft MQ series AS/400 and Oracle AQ as the interface between the WMS/ERP and the Vocollect voice application. Its TekRF middleware is certified by SAP for interfacing with SAP ERP systems.

The hardware

Two hardware suppliers dominate the warehousing and logistics market - Vocollect and Voxware. Other players in the market tend to focus on VR for call centres, or on applying text-to-speech technology in mobile applications and multi-language voice applications such as navigation systems.

Both Vocollect and Voxware now offer worker-management applications as well as simple hardware integration. Voxware's VoiceLogistics uses software techniques to monitor each worker and boost individual performance. It knows the experience and capability level or each employee, so it can balance workloads between workers in a team to boost performance of the whole group. It will permit expert workers to take shortcuts and provide extra coaching to trainees and temporary staff. You can add richer vocabularies and eliminate restrictions that tie specific workers to specific computer units.

At the heart of the system is Voxware's VLS-410 computer, which has a 400MHz processor, 128MB memory and 64Mb of flash memory. It works on any 802.11b wireless network. Battery life is 12 hours with a standard lithium ion battery. The VLS-410 uses VR2 Voice Recognition. There is no mute button, says Voxware, because the software automatically ignores out-of-vocabulary words.

Many of the basic components are similar to those you would find on a typical handheld terminal running some variant of Windows CE, but Voxware emphasises the platform-independence of its solution. It achieves this by using XML to transmit information in the form of a hidden "Web page", and something called VoiceXML (a Voxware patent) to make it all work. The data can include both a task-specific application, and work assignment data related to the individual picker's job for the day.

In effect, the VLS-410 incorporates a special form of XML browser, through which the VLS-410 communicates directly with the VoiceLogistics Voice Server. Voxware says this is preferable to using an independent voice server because it offers greater flexibility for tuning, management and control. Any worker can use any VLS-410 - there is no need to map specific devices to specific users.

Demonstrating the versatility of its approach, Voxware has teamed up with LXE, and now also offers voice recognition on LXE's MX3X Windows CE-based handheld computer, using VoiceLogistics software. LXE's MX3 and MX6 units are both audio-capable devices; the same unit can be used as a wearable computer for voice-directed logistics, or as a wireless handheld device (or indeed as a vehicle-mount computer).

Vocollect acknowledges that some terminals such as the MX3 are suitable for audio outputs, but says some claims that handheld RF terminals can also work with voice need treating with caution. "They will not necessarily have the full circuitry necessary to deliver optimum performance," a spokesman maintains. "Also voice and audio are both power-hungry applications, so check battery capability."

Vocollect's Talkman T2 is a dedicated voice terminal, whose underlying operating system is Windows Mobile in one configuration or another. The belt-worn unit itself is linked by a thin cable to the headset. The internal architecture supports 802.11b Wi-Fi compliant radios.

Vocollect's VoiceDirect synchronises warehouse activities in real time with the WMS, while VoiceLink provides managers with a view of voice-directed application data and individual operator activities on the Talkman wearable computer via a middleware connection. VoiceLink 2.0 expands the capability of the system to include put-away, replenishment, inventory transfers and line-loading.