Solutions / Catalogue / Acoustic & Audio AI

Acoustic & Audio AIScenario

On-Site Voice Control

A voice add-on for kiosks, desk robots and robot arms that hears people from a distance, lets them interrupt, and understands speech on a local box. After first setup it can run without internet. It is for makers of public-facing machines and firms on private networks that want to avoid per-use cloud charges.

RetailManufacturingRobotics and education

Talk to an engineer See the hardware

A cook seen from behind making masala dosas on a hot griddle in a busy restaurant kitchen in Mysuru — Photo: Sunnya343, CC BY-SA 4.0, Wikimedia Commons. Shows the setting, not this installation.

0.9 s: reply time (labelled measured, test conditions not given)
up to 80 dB: own playback the array can cancel (spec)
about 5 m: far-field pickup range (spec)
about 11 s: spoken command to object returned in the grasp demo (stated)

Results reported by the design's authors on their own test setups. Yours depend on the site, so we confirm them in a pilot.

The problem

Why it matters.

Voice devices hear their own playback, cloud LLM backends fail offline and bill per token, and cloud lag makes voice-driven robots unsafe.

The flow

From signal to action.

01SenseMic arrayA ReSpeaker array picks up speech, finds the speaker's direction and removes the device's own playback.
02DecideWake wordA small front end listens for a custom wake phrase in English or Chinese.
03DecideLocal ASRAn edge host turns speech into text in one of up to 30 languages.
04DecideUnderstandA local or cloud language model works out the intent and matches it to a command.
05ActRespondThe host speaks a reply, sends text to a business system, or sends motion commands to a robot.

Deployment architecture

How it is wired.

The devices on site, what carries their data, where the decision is made and where the result goes.

01SenseOn site

XIAO ESP32-S3 Sense1 per device
ReSpeaker Lite1 per device
ReSpeaker XVF3800
ReSpeaker Flex Circular-4
reSpeaker Clip

02ConnectCarries the data

reRouter CM4 1020321 per device
Bluetooth LE
Wi-Fi

03DecideRuns the model

reComputer RK3576 Series
reComputer RK3588 Series
reComputer J30 Series
reComputer J40 Series
reComputer J50 Series
NVIDIA Jetson AGX Thor Developer Kit

04Your systemsWhere results land

USB audio
I2S
Robot motion controller bus
Seeed solution CLI

05ActWhat happens next

Respond

How it works

Step by step.

01Pick a mic for the distance: near field up to 3 m or far field up to about 5 m.
02Pick a host by how many voices it must handle at once and whether it runs a local LLM.
03Deploy once with internet so images and models download, and set the language.
04Set the wake phrase, sensitivity and whether users can interrupt replies.
05Connect the output to your kiosk app, business system or robot controller.

The hardware

What goes on site.

Items we stock can go straight in the bag. Everything else we source and quote with the installation.

XIAO ESP32-S3 Sensewake word front end for your own host1 per device₹1,699
ReSpeaker Litetwo-mic near-field pickup1 per deviceSourced on quote
reRouter CM4 102032entry voice host1 per deviceSourced on quote
ReSpeaker XVF3800four-mic far-field array with echo cancellationSourced on quote
ReSpeaker Flex Circular-4360 degree pickup with 10 W amplifierSourced on quote
ReSpeaker Flex Linear-4180 degree pickup with 10 W amplifierSourced on quote
reSpeaker Clipwearable pickupSourced on quote
reComputer RK3576 Seriesvoice host, one channelSourced on quote
reComputer RK3588 Seriesvoice host, one channelSourced on quote
reComputer J30 Seriesvoice host with natural voiceSourced on quote
reComputer J40 Seriesvoice host, 2 to 3 channels, 4B to 7B LLMSourced on quote
reComputer J50 Seriesvoice host, 3 or more channelsSourced on quote
NVIDIA Jetson AGX Thor Developer Kittop host, 6 or more voice channels₹3,39,000

Before you install

Internet for the first start to download images and models.
A mic and speaker position within the array's rated pickup range of 3 to 5 m.
A host matched to the number of simultaneous conversations.
Quiet enough surroundings, since motor and gear noise is not cancelled.

Good to know

Running a local LLM takes compute from the voice pipeline, so plan for one conversation at a time.
Echo cancellation removes the device's own audio only, not machine noise.
The robot grasp demo cannot handle clear objects or items wider than 10 cm.
With a cloud model, recognised text leaves the site even though audio does not.

How we deliver it

Paper to site, one team.

01Site surveyWe walk the site, or review photos and drawings, and agree what to measure and where each device goes.
02PilotOne lane, one line or one room first. We tune it on your conditions and agree what good looks like.
03Roll outWe supply the hardware in India with GST invoices, install it, and connect it to the systems you already run.
04SupportSpares, remote checks and model updates, with one team to call when something changes.

Questions

Asked before buying.

Does it work offline?

Yes, after a first online start downloads the models. A cloud LLM option needs network.

Can people interrupt the machine while it is talking?

Yes. Speech during a reply stops playback, and this can be switched off.

How many people can talk to one box at once?

From one channel on entry hosts up to six or more on Jetson AGX Thor, and one channel if a local LLM is running.

Want this on your site?

Tell us the site and the problem. We come back with a pilot plan and a priced hardware list.

Book a call with an engineer Write to us

Want this on your site?

In-Person Conversation Intelligence

Counter Audio Capture & Analytics

AI Voice Warehouse Management

Want this on your site?

In-Person Conversation Intelligence

Counter Audio Capture & Analytics

AI Voice Warehouse Management