Voice Directed Picking System: How Pick-by-Voice Works
A voice directed picking system reads order locations and quantities into a picker's headset and has them speak check-digits back to confirm — hands-free, eyes-free, no paper sheet. Here's plainly how pick-by-voice works, its honest pros and cons, when it's actually worth it, and the part nobody sells you: voice is only as accurate as the stock system feeding it.
A voice directed picking system gives warehouse pickers their instructions through a headset and takes their confirmations by speech. The system speaks a location, the picker reads back a short check digit from the shelf label, the system speaks a quantity and the picker confirms it, all without a paper list, a screen or a handheld scanner.
The practical problem is that voice is usually bought as a device, when it behaves more like a conversation between the picker and the warehouse management system (WMS). The dialogue, the check digits, the hardware, the WMS interface and the rollout plan each decide whether that conversation stays fast and accurate. This guide explains how each part works and what to check before you commit.

A paper list occupies a hand and the picker’s eyes; voice moves the same instructions into a headset. Illustrative photograph, not an OpsMavix project.
Image credit: Centre for Ageing Better, Unsplash License. No edits.
Key Takeaways
- Voice picking is a spoken dialogue: the system gives a location and quantity, and the picker confirms each with a short response.
- A spoken check digit confirms the location, not the identity of the item in it.
- Check digits need rules: no repeats at neighbouring slots and periodic changes so pickers cannot recite them from memory.
- The headset and wearable device are the visible part; wireless coverage, chargers and spares decide uptime.
- The WMS interface must handle short picks, lost connections and reassigned work, not just normal lines.
- Pilot one zone with your own staff, then expand in phases using measured results.
In this guide:
- 1. Understand What a Voice Directed Picking System Does
- 2. Follow the Voice Dialogue for One Pick
- 3. Design Check Digits That Stay Reliable
- 4. Choose the Headsets and Voice Devices
- 5. Connect the Voice System to Your WMS
- 6. Plan Speech Recognition and Training
- 7. Compare the System Approaches
- 8. Implement Voice in Phases
- Frequently Asked Questions
- How OpsMavix Can Help
1. Understand What a Voice Directed Picking System Does
The four parts of the system
A voice system has four layers. Each one can fail independently, so it helps to name them before talking to suppliers.
| Layer | What it does | Typical owner |
|---|---|---|
| Headset | Speaker and microphone worn by the picker | Operations, with hygiene and replacement rules |
| Voice device | Wearable or handheld computer running the voice application | IT or the voice supplier |
| Voice application or middleware | Turns WMS tasks into spoken prompts and turns responses into confirmations | Voice supplier or integrator |
| WMS or order system | Holds orders, locations, stock and the record of what was picked | Warehouse systems owner |
Where voice sits among picking decisions
Voice is a picking technology. It decides how instructions arrive and how picks are confirmed. It does not decide whether orders are picked one at a time, in batches or by zone; that is a picking strategy, covered in our guide to picking strategies compared. If you are choosing between voice alone and voice combined with scanning and a screen, read multi-modal vs voice-directed picking. This guide stays with how the voice system itself works.
What voice is good at
Honeywell’s support documentation for its Talkman voice devices states that “All Talkman devices leave the operator’s hands free to inspect items, pick products, drive vehicles, or repair defects.” That is the core benefit: both hands stay on the product and both eyes stay on the shelf and the travel path. It suits case picking, heavy items, cold stores where gloves make screens awkward and long runs of simple lines.
2. Follow the Voice Dialogue for One Pick
A typical exchange
Every vendor’s dialogue differs in wording, but most follow the same pattern of prompt and response. A simplified example:
| Step | System says | Picker says | What the system checks |
|---|---|---|---|
| 1 | “Aisle 14, slot 22.” | (walks to the location) | Nothing yet |
| 2 | (waits) | “Five three.” | Spoken digits match the check digit stored for that slot |
| 3 | “Pick 4.” | “4.” | Quantity matches, or a short pick is declared |
| 4 | “Put in tote 2.” | “Ready.” | Item is assigned to the correct order container |
| 5 | “Aisle 14, slot 31.” | (walks on) | Next task in sequence |
If the spoken digits do not match, the system repeats the location rather than accepting the pick. That single rule is what turns voice from spoken instructions into a confirmation workflow.
Words the picker must learn
Beyond numbers, pickers use a small vocabulary of commands. Typical examples include asking the system to repeat a prompt, reporting a short pick, skipping a location that cannot be reached and asking how many lines remain. Agree this vocabulary with the supplier early, because every extra command is something new starters must learn.
Hypothetical worked example: why prompt length matters
Assume a picker completes 600 lines per shift and each line involves these spoken steps:
- Location prompt: 2.0 seconds.
- Check digit response: 1.0 second.
- Quantity prompt: 1.0 second.
- Quantity response: 1.0 second.
Talk time per line is 2.0 + 1.0 + 1.0 + 1.0 = 5.0 seconds. Across the shift that is 600 × 5.0 = 3,000 seconds, or 50 minutes.
Now assume the location prompt is shortened by 0.5 seconds, for example by dropping words experienced pickers no longer need. The saving is 600 × 0.5 = 300 seconds, or 5 minutes per picker per shift. For a hypothetical team of 20 pickers working 250 shifts a year, that is 20 × 5 × 250 = 25,000 minutes, or about 416.7 hours a year.
All figures are hypothetical. Measure your own prompt and response times during a pilot, and ask whether the supplier lets you adjust prompt wording and speed by user experience level.
3. Design Check Digits That Stay Reliable
What a check digit proves
A check digit is a short number printed on each location label that the picker speaks on arrival. It confirms the picker reached the intended location. It does not confirm the right product is in that location; if stock was put away in the wrong slot, the check digit still matches. That is why accurate warehouse bin locations and reliable put-away matter as much as the voice system.
Two known weaknesses
A Vocollect patent on generating and updating location check digits, granted in 2014, sets out two problems. First, pickers can memorise check digits through repetition and speak them before reaching the slot. Second, randomly assigned digits can repeat at adjacent or related slots, which raises the risk of a wrong-location pick being accepted. The patent describes excluding digits used at neighbouring locations and at the same slot number in other aisles when choosing new ones.
Lucas Systems describes the same memorisation issue in its voice versus RF discussion, saying “To prevent memorization, the solution is to periodically change the checkstrings,” and that its Jennifer software lets managers change them “on a daily basis (or more) without changing the location labels.” Ask each supplier how it handles rotation and what that means for your labels.
Hypothetical worked example: two or three digits
Assume each check digit is assigned at random, ignoring exclusion rules, and that in a year a hypothetical warehouse has 400 occasions where a picker arrives at the wrong slot and reads that slot’s label.
- With two digits there are 100 possible values (00 to 99), so the chance the wrong slot shares the correct slot’s digits is 1 in 100, or 1%. Expected undetected arrivals: 400 × 1% = 4.
- With three digits there are 1,000 possible values, so the chance is 1 in 1,000, or 0.1%. Expected undetected arrivals: 400 × 0.1% = 0.4.
Excluding repeats at adjacent slots reduces the risk further for the most likely wrong slots, because those are the ones a picker is most likely to reach by mistake. Longer digits add a little talk time to every line, so the trade-off is accuracy against speed.
Check digit design checklist
- Decide the digit length per zone, not just for the whole building.
- Exclude repeats at adjacent slots and at the same slot number in neighbouring aisles.
- Agree how often digits change and who approves the change.
- Print labels that can be read at the picking height and in your lighting.
- Keep the location master and the labels in step when locations are added or moved.
4. Choose the Headsets and Voice Devices
Dedicated voice devices and general wearables
There are two broad hardware routes. Dedicated voice devices are built for one job. Honeywell’s Voice A720x product page lists an IP67 rating, Bluetooth, two connectors for wired headsets and scanners, no built-in scanner, and an operating range of minus 30°C to 50°C with a cold storage option.
General wearable computers can also run voice applications alongside other apps. Zebra’s WT6300 specification lists wired and Bluetooth headset support and corded or Bluetooth ring scanners. Whether a particular voice application is supported on a particular device is something to confirm with the software supplier, not assume.
| Factor | Dedicated voice device | General wearable or handheld |
|---|---|---|
| Main purpose | Voice workflows | Several apps, including voice |
| Screen | Usually none | Usually yes |
| Scanning | Through an added scanner, if supported | Integrated or paired ring scanner |
| Software choice | Tied to the voice platform | Depends on the voice supplier’s supported list |
| Best fit | Voice-only zones, cold stores | Sites mixing voice with scanning or screens |
Hypothetical worked example: sizing the kit
Assume a hypothetical site with 18 pickers on the day shift and 12 on the late shift, sharing devices between shifts but issuing personal headsets for hygiene.
- Devices needed at peak: 18. Adding 10% spares: 18 × 1.1 = 19.8, rounded up to 20 devices.
- Personal headsets: 18 + 12 = 30, plus 2 spares = 32 headsets.
Charging capacity, battery swaps and repair turnaround should be confirmed with the supplier against your actual shift pattern.
Hardware checklist
- Wireless coverage reaches every pick location, including mezzanines and cold stores.
- Chargers are sized for devices and batteries returning at the same shift change.
- Spare devices cover faults and new starters.
- Headsets suit your noise levels and any protective equipment pickers wear.
- Someone owns repairs, replacements and device updates.

Every location a headset names must exist, with the right stock, in the WMS. Illustrative photograph, not an OpsMavix project.
Image credit: Arum Visuals, Unsplash License. No edits.
5. Connect the Voice System to Your WMS
The message flow
The voice application does not decide what to pick. It receives tasks, speaks them and returns results. A typical sequence looks like this:
- The WMS releases a batch of pick tasks with location, item, quantity and container.
- The voice layer adds the check digit for each location and sequences the prompts.
- The picker confirms each line; the voice layer sends the confirmed quantity back.
- Shorts, skips and damaged items are sent back as exceptions.
- The WMS updates stock, triggers replenishment and releases the order to packing.
For a broader view of how stock and order systems exchange records, see our guide to ERP and WMS integration.
Decisions to agree with the supplier
| Decision | Options | Why it matters |
|---|---|---|
| Confirmation timing | Each line in real time, or in groups | Real-time updates keep stock and replenishment current |
| Lost connection | Device holds work offline, or picking pauses | Offline work must not create duplicate confirmations |
| Short picks | Spoken short with a reason code | Must trigger replenishment and stock investigation |
| Work reassignment | Supervisor moves tasks between pickers | Prevents two pickers being sent for the same line |
| Check digit source | Held in the WMS or in the voice layer | Decides who updates digits when locations change |
Test the exceptions, not only the normal pick
A normal line proves only one path works. Test a short pick, a location with the wrong stock, a lost connection mid-batch, a task reassigned between pickers and a location renamed in the WMS. If any of these leaves stock or orders out of step, the voice system will speak wrong data with complete confidence. Reliable pick face replenishment also matters, because a short pick at an empty face is a replenishment failure, not a voice failure.
6. Plan Speech Recognition and Training
Recognition, accents and noise
Recognition quality depends on the software, the headset, the noise on your floor and the voices of your team. Confirm which languages are supported for prompts and responses. Suppliers take different approaches, including whether each user records a voice profile before starting. Rather than rely on general claims, test recognition with new starters, experienced pickers and staff whose first language differs from the system default, in your noisiest zone at your busiest time.
Training plan
| Stage | What the picker learns | How to check it |
|---|---|---|
| Before the floor | Headset fit, vocabulary, voice profile if required | Completes a practice dialogue without help |
| First shifts | Normal picks, repeat requests, short picks | Supervisor observation and exception log |
| After a week | Full vocabulary and pace | Stable lines per hour and error rate |
Training checklist
- Pair new starters with an experienced voice picker for the first shift.
- Log misrecognitions by user and zone to spot headset or noise problems.
- Agree who maintains user profiles, prompts and vocabulary after go-live.
7. Compare the System Approaches
Match the approach to the gap
Voice can be added in several ways. The right one depends on what your current systems already do and where the gap is.
| Approach | Where it fits | What to check |
|---|---|---|
| Configure current tools | Your WMS already offers voice or supports a voice partner | Licence edition, supported devices and configuration effort |
| Connect existing systems | A packaged voice product sits beside your current WMS or order system | Interface method, exception handling and who supports it |
| Packaged voice software with a WMS | Standard workflows fit and you want an established product | Supported connectors, hardware list and support terms |
| Focused custom application | One clear gap, such as voice confirmation feeding a custom stock system | Data access, maintenance and integration scope |
| Full ERP, packaged or custom | Picking is one of several connected problems across stock, orders and finance | Fit across functions, migration and delivery capacity |
Packaged voice products are mature and often the sensible choice where the workflows fit. Ask every option to demonstrate the same day of real order lines and three exceptions, and compare hardware, licences, integration and support, not just device prices. A focused build makes most sense when your stock and order records already live in a custom system, or when the dialogue you need does not fit the package without heavy workarounds.
8. Implement Voice in Phases
Prepare the data first
Before any headset reaches the floor, confirm that every pilot location exists in the WMS, carries the correct stock and has a printed check digit label. Clean item data, location data and replenishment rules shorten the pilot more than any training session.
Run a measured pilot
Choose one zone and one order type. Record a baseline week with the current method, then run voice for at least the same period using the same measures: lines per hour, errors by type, time for new starters to reach a stable rate, misrecognitions and integration exceptions. Keep the picking strategy and slotting unchanged so voice is the only variable.
Hypothetical worked example: phasing a rollout
Assume a hypothetical site with 2,400 pick locations across 4 zones of 600 locations each.
- Phase 1: pilot zone, 600 labels, 6 pickers.
- Phase 2: two further zones, 1,200 labels.
- Phase 3: final zone, 600 labels.
Total labels: 600 + 1,200 + 600 = 2,400. If labels are reprinted once a year for rotation or damage, that is another 2,400 labels annually, unless the supplier can rotate digits without relabelling.
Rollout checklist
- Pilot locations and stock are verified in the WMS.
- Check digit labels are printed, fitted and tested.
- Wireless coverage is surveyed across the pilot zone.
- The interface handles tasks, confirmations, shorts and lost connections.
- Recognition is tested with your real pickers in your noisiest zone.
- Devices, headsets, chargers and spares are in place.
- Supervisors can see exceptions during the shift.
- Pickers give feedback before the next phase starts.
- Results are compared with the baseline week, not with supplier estimates.
- Each phase is approved before the next zone is labelled.
Frequently Asked Questions
What is a voice directed picking system?
It is a picking method where a headset gives the picker spoken instructions from the warehouse system, and the picker confirms locations and quantities by speech. It keeps hands and eyes free and removes paper pick lists.
How do check digits work?
Each location label carries a short number. On arrival the picker speaks it, and the system accepts the pick only if it matches the number stored for that location. It confirms the location, not the item.
Why do check digits need to change?
Pickers can memorise them through repetition and recite them without reaching the slot. Suppliers address this by changing digits periodically and avoiding repeats at neighbouring locations.
Can voice picking confirm the actual product?
Not with a location check digit alone. Some systems let pickers speak the last few digits of a product code or add an occasional barcode scan, which is where multi-modal picking comes in.
What hardware does voice picking need?
A headset, a wearable or handheld device running the voice application, wireless coverage across the picking area, check digit labels on locations, and chargers and spares for every shift.
Does voice picking need a WMS?
It needs a system that issues pick tasks and accepts confirmations and exceptions. That is usually a WMS, but it can be an order management system or a focused application connected through the supplier’s interface.
Will accents or noise stop voice picking working?
They are factors to test rather than assume. Trial recognition with your own team, in your loudest zone, during the busiest period, before committing to a full rollout.
Can OpsMavix build a voice directed picking system?
Yes. OpsMavix can build focused applications or a full custom ERP, and can configure or integrate suitable existing tools where those approaches fit the requirements.
How OpsMavix Can Help
OpsMavix can map how pick tasks, confirmations and exceptions move between your voice devices and your stock and order records, then define where configuration, integration or a focused build closes the gap.
That may mean connecting a packaged voice product to an inventory automation system, tightening the order flow with a wholesale order management system or keeping channel stock aligned through ecommerce inventory automation.
Contact OpsMavix with one recent shift of picking exceptions, the records behind it and the handoff that creates the most rework. We can use that evidence to define requirements and a practical pilot.
Getting value from OpsMavix? Add us as a preferred source on Google — you'll see more of our operations content in your AI Overviews, AI Mode and Search.