Skip to main content

Smart Retail Voice AI

Leverage Seeed's Smart Retail Voice AI solution to capture and analyze in-store customer conversations. Our edge AI technology enables on-device speech-to-text processing, transforming voice data into actionable insights without relying on cloud services. Enhance customer experience, optimize store operations, and gain valuable feedback with our offline AI-powered voice analytics tailored for retail environments.

Smart Retail Voice AI
Solution Bundle 🖱️

Smart Retail Voice AI

Build an in-store voice analytics pipeline using edge devices and SenseCraft Voice to capture conversations on-site, run on-device speech-to-text, and turn customer voice into structured data.

  • On-Device Speech-to-Text at the Edge

    SenseCraft Voice runs directly on in-store edge devices to handle voice activity detection and speech recognition locally. All speech is processed at the edge, dramatically reducing dependency on cloud connectivity for real-time transcript generation.

  • Far-Field, Retail-Ready Audio Capture

    The solution is designed for far-field audio capture in noisy retail environments. With support for microphone arrays, beamforming, and noise suppression, it focuses on real customer–staff conversations even with background music and surrounding chatter.

  • Privacy-First Edge Architecture

    Because audio is transcribed locally on edge devices and only text plus metadata are sent upstream, retailers can better control privacy and compliance risks, while keeping cloud costs and network requirements low.

  • From Conversations to Data-Driven Decisions

    Transcripts generated by SenseCraft Voice become structured, analyzable data. Retailers can measure service quality, identify common questions and pain points, and continuously improve scripts, policies, and overall store experience.

Getting Started​

In this section, we will guide you through the steps to set up the Smart Retail Voice AI solution using Seeed's reRouter and reSpeaker XVF3800 microphone array.

1. Hardware Requirements​

This guide uses Seeed's reRouter and the reSpeaker XVF3800 microphone array to demonstrate how to build a Smart Retail Voice AI solution.

reRouter (Edge Compute)reSpeaker XVF3800 (Audio Capture)
reRouter CM4reSpeaker XVF3800

Role: Processing Unit

Technical Specifications:

  • RAM: 4GB (Minimum Requirement)
  • Storage: 32GB eMMC
*Note: Performance below these specs has not been validated.

Role: Voice Capture

Technical Specifications:

  • Model: XVF3800 4-Mic Array
  • Features: Noise Suppression, AEC, Beamforming

Hardware Setup​

product-rerouter-respeaker

1.1 Get reRouter Firmware​

There are two versions of the OpenWrt firmware available for reRouter:
one for global users, and another optimized for users in Mainland China.

The current firmware is based on OpenWrt 24.10.3 (Build r28872).

Get the firmware🖱️

Please refer to the reRouter flashing guide for detailed instructions on the flashing procedure.

Note that: you should use the firmware provided above, not the one from the guide page.

1.2. Hardware Connection​

Please connect the reRouter as follows:

  • Internet Access: Connect an Ethernet cable from your router/modem to the WAN port.
  • Host Connection: Connect wirelessly to the Wi-Fi hotspot named OpenWrt-XXXX or connect your computer to the LAN port via Ethernet.
reRouter hardware connections: LAN port connects to your host computer for configuration; WAN port connects to the internet.
  • WAN Port: Connects to the internet.
  • LAN Port: Connects to your host computer for configuration.

1.3. reSpeaker XVF3800 Setup​

Connect the reSpeaker XVF3800 microphone array to your host device via USB to configure.

  1. Clone the repository and navigate to the control folder:

Depending on your device, replace <YOUR_HOST_DIR> with the folder matching your system (e.g., linux_x86_64, rpi_64bit, mac_arm64, or jetson).

git clone https://github.com/respeaker/reSpeaker_XVF3800_USB_4MIC_ARRAY.git
cd reSpeaker_XVF3800_USB_4MIC_ARRAY/host_control/<YOUR_HOST_DIR>
  1. Execute the configuration sequence:

Grant execution permissions and run the following three commands to initialize the device (use sudo if required on Linux):

chmod +x ./xvf_host

# 1. Clear existing configuration
sudo ./xvf_host clear_configuration 1

# 2. Enable specific audio manager setting
sudo ./xvf_host audio_mgr_op_r 8 0

# 3. Save configuration
sudo ./xvf_host save_configuration 1

2. Software Installation​

2.1. Prerequisites​

Before starting, ensure the following conditions are met:

  • Access: Root access to your OpenWrt device via SSH (e.g., ssh [email protected]).
  • Network: The OpenWrt device has a stable internet connection.
  • Audio: The reSpeaker XVF3800 is connected to reRouter.

2.2 Accessing the Device​

Once your computer is connected to the reRouter (via LAN or Wi-Fi), you can access the system using the following defaults:

  • SSH Access:
ssh [email protected]

By default, no password is set for the root user.

Note

You can also use the OpenWrt web interface to configure network settings, including connecting the reRouter to your Wi-Fi or Ethernet network for internet access.

  • Open a browser and visit: http://192.168.49.1
  • Username: root
  • Password: (none / empty by default)

If you need other language support, such as Chinese, you can install the luci-i18n-base-zh-cn package via the web interface or SSH.

opkg update
opkg install luci-i18n-base-zh-cn
Important

Before proceeding to the installation steps below, please verify that the reRouter can access the internet (for example, run ping google.com or ping openwrt.org in the SSH terminal).

2.3 Step-by-Step Installation​

Execute the following commands sequentially after establishing an SSH connection to your OpenWrt device.

Step 1: Install Docker and Dependencies​

This step installs the container runtime environment and necessary tools, including the SHA-256 checksum utility for file verification.

# 1. Update the local package list
opkg update

# 2. Install Docker core components
# Note: On some systems, you might need to install these separately: dockerd, docker, containerd, runc
opkg install dockerd docker containerd runc

# 3. Install utility packages
opkg install wget-ssl unzip ca-certificates

# 4. Enable and start the Docker daemon service
/etc/init.d/dockerd enable
/etc/init.d/dockerd start

# Optional: verify downloaded files
# sha256sum <filename>
Step 2.2: Prepare Data Directories and Configuration​

Create the persistent storage directories required by the SenseCraft containers and download the default configuration file.

# 1. Create required application data directories
mkdir -p /data-iot/respeaker/recordings \
         /data-iot/respeaker/models \
         /data-iot/respeaker/voiceprints \
         /data-iot/respeaker/logs

# 2. Navigate to the base directory
cd /data-iot/respeaker

# 3. Download the configuration file
wget -q -O config.yaml 'https://appstore.seeed-fleet.com/config.yaml'
Step 2.3: Download and Extract Models​

We will download the pre-trained ASR model package directly from the Seeed Studio server, verify its integrity using SHA-256, and extract it.

FileURL
models.ziphttps://files.seeedstudio.com/wiki/solution/ai-sound/reRouter-firmware-backup/models.zip
Expected SHA-256 Hash7b9e7606a2ddcad56f3f72a77b16eb2c60437ae4bfc3f1423bd33db177385c9d

Execute the following commands on the OpenWrt host via SSH:

# 1. Navigate to the base directory
cd /data-iot/respeaker

# 2. Define the Model URL
MODEL_URL="https://files.seeedstudio.com/wiki/solution/ai-sound/reRouter-firmware-backup/models.zip"

# 3. Download the large model package. 
# -c allows the download to resume if interrupted.
wget -O models.zip -c "$MODEL_URL"
echo "Model package download completed. Check file size is approximately 500MB."

# 4. Verify the file integrity using SHA-256 Checksum
# The result MUST match the expected hash above.
sha256sum models.zip

# 5. Extract the model package into the 'models' directory
unzip -o models.zip

# 6. Clean up the temporary ZIP file
# rm -f models.zip

# 7. Verify the model files are present
ls -l /data-iot/respeaker/models/
Step 2.4: Configure Audio Device Permissions​

The Voice Client container requires privileged access to the audio devices. Ensure the device files have appropriate permissions.

# 1. Check for audio devices
ls -l /dev/snd/

# 2. Set read/write permissions for all users on audio device nodes
chmod -R 666 /dev/snd/*
Step 2.5: Deploy Docker Containers​

Define the image variables and launch the three required services: sensecraft-voice-client, sensecraft-asr-server, and watchtower.

# Define Image Variables
VOICE_CLIENT_IMAGE="sensecraft-missionpack.seeed.cn/respeaker/sensecraft-voice-client:v0.0.2"
ASR_SERVER_IMAGE="sensecraft-missionpack.seeed.cn/respeaker/sensecraft-asr-server:latest"
WATCHTOWER_IMAGE="sensecraft-missionpack.seeed.cn/respeaker/containrrr/watchtower:latest"

# 1. Pull Images
docker pull $VOICE_CLIENT_IMAGE
docker pull $ASR_SERVER_IMAGE
docker pull $WATCHTOWER_IMAGE

# 2. Stop and remove any containers with the same names
docker rm -f sensecraft-voice-client sensecraft-asr-server watchtower

# 3. Launch sensecraft-voice-client 
docker run -d --name sensecraft-voice-client \
    --restart=unless-stopped \
    --privileged \
    --device /dev/snd \
    --network host \
    --group-add audio \
    --group-add video \
    --cap-add SYS_ADMIN \
    --cap-add SYS_RAWIO \
    -e TZ=Asia/Shanghai \
    -e AUDIO_CARD_ID=1 \
    -e AUDIO_DEVICE_ID=0 \
    -v /etc/wpa_supplicant:/etc/wpa_supplicant \
    -v /etc/network:/etc/network \
    -v /var/run/dbus:/var/run/dbus \
    -v /dev:/dev \
    -v /run/udev:/run/udev:ro \
    -v /proc:/proc:ro \
    -v /sys:/sys:ro \
    -v /data-iot/respeaker/recordings:/app/recordings \
    -v /data-iot/respeaker/voiceprints:/app/voiceprints \
    -v /data-iot/respeaker/logs:/app/logs \
    -e WIFI_INTERFACE=wlan0 \
    -e WIFI_CONFIG_PATH=/etc/wpa_supplicant \
    $VOICE_CLIENT_IMAGE

# 4. Launch sensecraft-asr-server (model server)
docker run -d --network host \
    -v /data-iot/respeaker/models:/app/models \
    -v /data-iot/respeaker/voiceprints:/app/data \
    --restart=always \
    --name=sensecraft-asr-server \
    $ASR_SERVER_IMAGE

# 5. Launch watchtower (for continuous container monitoring and update)
docker run -d --name watchtower \
    --restart always \
    -v /var/run/docker.sock:/var/run/docker.sock \
    $WATCHTOWER_IMAGE \
    --cleanup -i 60 sensecraft-asr-server sensecraft-voice-client
3. Verification​

Check the final state of the deployment.

# Check container status (All three should show Status: Up)
docker ps

# Check the voice client logs for successful initialization and audio device detection
docker logs sensecraft-voice-client

If the logs indicate successful starting without critical errors, the SenseCraft services are successfully deployed.

it is highly recommended to reboot the device to ensure all settings, permissions, and network configurations are fully loaded and recognized by the system.

reboot

After reboot, you navigate to http://192.168.49.1:8090 to access the Edge Client interface for real-time ASR transcription and device configuration. For detailed usage of the SenseCraft Voice platform, please refer to the User Guide section below.

SenseCraft Voice: Edge-to-Cloud Platform Overview​

SenseCraft Voice is a cutting-edge platform designed to transform raw audio data captured at the edge (reRouter) into actionable business intelligence through powerful AI analysis and centralized management.

The platform's unique edge-cloud architecture provides unparalleled reliability, speed, and analytical depth for enterprise-grade audio monitoring solutions.

FeatureValue PropositionKey Advantages
Resilient Edge ProcessingGuarantees continuous operation and low latency.Voice ASR and recognition run locally on the reRouter, ensuring real-time response and data collection even during network outages.
Deep AI CustomizationAdapts the platform to specific business needs and terminology.Administrators can define custom Keywords, Synonyms, and AI Prompts to direct AI analysis and ensure accurate event detection specific to their business language.
Granular Location MappingSimplifies large-scale deployment management.Supports hierarchical organization of thousands of edge devices by Store, Location, and Device Name, moving beyond confusing MAC addresses for easy filtering and reporting.
Actionable DashboardsProvides immediate business insight and performance tracking.Centralized Dashboard features Multi-Store Filtering, real-time Device Online Rate, and Keyword Hotspot Analysis to monitor operational status and business events instantly.

The SenseCraft Voice solution is built upon a robust edge-cloud architecture, ensuring both real-time local processing and centralized management. The service consists of two primary components: the Edge-side Client running on the reRouter, and the Cloud/Server-side Management Platform.

User Guide​

Edge-side Client (reRouter) Access​

The Edge Client is essential for real-time validation and local setup.

  • Access: Open your web browser and navigate to the reRouter's IP address on port 8090: http://192.168.49.1:8090.
  • Core Function: The interface provides real-time ASR transcription (to verify audio input), controls for Voiceprint Recognition (speaker identification), and Device Configuration (network settings, upstream server address).
Module Name Description Interface Screenshot
Voice ASR

Description: Displays the current operational status of the local Automatic Speech Recognition (ASR) service.

Purpose: Provides real-time transcription of detected speech, essential for verifying local audio input and recognition accuracy.
Voice ASR Module Interface

Figure 1: Voice ASR Module
Voiceprint Recognition

Description: Manages and monitors the Voiceprint Recognition system.

Purpose: Automatically generates unique voiceprints from audio recordings to enable speaker differentiation and identification.
Voiceprint Recognition Module Interface

Figure 2: Voiceprint Recognition Module
Device Status & Configuration

Description: Provides detailed information about the reRouter's operating status and allows core parameter changes.

Purpose: Enables configuration updates such as network settings (Wi-Fi) and changing the upstream server address for cloud communication.
Device Status and Configuration Module Interface

Figure 3: Device Status & Configuration

Cloud-side Management Platform​

The Cloud Platform is organized into five main navigation areas, offering powerful data analysis and system configuration tools.

1. Dashboard: Insights at a Glance​

The Dashboard is your operational command center, providing aggregated metrics and performance trends:

  • Store Filtering: Easily switch views by selecting one or more stores, with all charts updating instantly.
  • Analysis: Monitor Daily Collection Trends (hourly records) and Keyword Hotspot Analysis (showing which keywords are frequently triggered and the associated Device Names).
Dashboard Main Page
Figure 4: Dashboard Interface

2. Record Management: Data Auditing & Export​

This module provides the definitive view of all collected voice records.

  • Advanced Filtering: Use Device Name, Store Name, Location Name, or MAC Address for precise data retrieval. Searches are executed only after clicking the "Filter" button, giving users full control.
  • Export Capability: Select and export filtered data in three formats for external use (choose one at a time): Markdown, Plain Text (.txt), or Original Audio File.
  • Dual-View Auditing: Easily switch between Conversation Mode to review transcribed dialogues and Timeline Mode to listen to the original audio playback. This dual approach allows for rapid verification of transcription accuracy and a deeper understanding of the interaction's context.
  • Clarity: All record views prioritize the easily identifiable Device Name over the MAC address.
Record Management Interface
Figure 5: Record Management Interface
sensecraft-voice-record
Figure 5.1: Timeline Mode

3. AI Analysis: Historical & Custom Processing​

This area handles the submission of speech records for advanced processing by the AI engine.

  • Historical Sessions: Review your past interactions with the AI analysis engine. The history window displays conversations chronologically, and clicking a session immediately loads the previous conversation thread for review.
  • Processing: Submit filtered records for AI processing based on the currently selected AI Prompt.
AI Analysis Interface
Figure 6: AI Analysis Interface

4. Store Management: Device & Location Hierarchy​

This area provides the necessary tools for setting up and maintaining the organizational hierarchy of all edge devices.

  • Hierarchical View: Easily manage stores, their specific in-store locations, and the associated reRouter devices.
  • Centralized Control: Streamline device deployment and configuration by grouping devices logically.
Store Management Interface
Figure 7: Store Management Interface
Device Management Interface
Figure 8: Device Management Interface
Specific Location Management Interface
Figure 9: Specific Location Management Interface

5. Backend Configuration: System Control & Customization​

This section allows administrators to define system-wide parameters for AI processing and event triggering.

5.1. Keyword Settings​

Define custom keywords and synonyms to identify specific business events in recordings.

  • Customization: Define Keywords and their Synonyms for event detection.
  • Visualization: Assign a Marking Color for visual distinction on the Dashboard.
  • Management: Supports Adding, Editing, Deleting, and Batch Deletion.
Keywords Backend Management Interface
Figure 10: Keyword Settings Interface
5.3. User Management​

The User Management module controls platform access and permissions.

User Management Interface
Figure 12: User Management Interface
5.2. AI Prompt Settings​

Create and manage custom AI Prompts to dictate how the AI processes selected voice records.

  • Control: Define prompt Name, Tags, and Content. Only one Enabled prompt is active for use at a time.
  • Management: Supports Adding, Editing, Deleting, and Batch Deletion.
AI Prompt Editing Interface
Figure 11: AI Prompt Settings Interface
Loading Comments...