ScenarioOn-Site Voice Control
A voice entry point users can interrupt, added to existing equipment; recognition runs on a local host and works offline after setup.
Acoustic & Audio AI
Hear clearly, then turn speech into text, voice and dialogue locally
Capture clear audio with AI microphones, then transcribe, synthesize speech or run an LLM on a local box. Open reference designs work out of the box.
Clear Audio
Clear speech in noisy places, with who spoke and what was said
Local Compute
Transcribe fully on-site and send only text
Tuned, Ready to Run
Tuned and benchmarked per board, open source to modify
Where this technology lands
Scenario
Reference designInterruptible Conversational Voice AI
A voice terminal that answers on the device and can be interrupted mid-answer. Listening and speaking stay local; the conversation model is either your own cloud endpoint or on-device. Runs on a reComputer RK, a reComputer J30 / J40 (Orin) or a reComputer R2000 series, with the conversation language fixed at deploy time.
ScenarioIn-Store & Counter Service QA
Counter conversations are turned into text on site, with the audio kept in the store. For staff who wear a recorder, see In-Person Conversation Intelligence.
ScenarioAI Voice Warehouse Management
The Watcher understands spoken commands and identifies the operator, writing results to the built-in warehouse system or your existing ERP / WMS. Deploy in the cloud, on a local host or fully offline.
Reference designRetail Voice
Store, meeting, consultation speech becomes back-office text. Mic array and on-site transcriber, or reSpeaker Clip and phone. Names, numbers masked before storage, deletable on request.
Reference designConversational Desktop Robot
Turn a desktop robot into a voice companion — listens, thinks, speaks, and expresses emotions in under 1 second. Works with Reachy Mini. Three deployment options: Jetson all-in-one, reComputer AI Industrial R21 with Hailo-8 NPU, or CM4 wireless.
Reference designSmart Warehouse Management
Voice-controlled warehouse management on edge hardware — speak the transaction, MCP writes it to inventory. Five tiers, from cloud trial to a fully offline Jetson deployment.
ScenarioBuilding Energy Retrofit
LoRaWAN sensors mount on the wall, one gateway connects existing Modbus and BACnet equipment, and the data goes to an open-source dashboard or your own platform.
ScenarioBuilding Energy Management
Connect existing and new devices by floor and system. Start with energy monitoring, then expand to control; view electricity, water, and HVAC data together.
Hardware
Transcription needs only an entry compute box; a local LLM needs RK3588 with an NPU card or a Jetson. Adding an LLM does not change the microphone.
Not sure which to choose? See the full selection guide →
Device Task
What does the device need to do?
AI microphone → compute box → speaker / output → application
The microphone picks up and cleans the audio and connects in one of three ways: USB firmware plugs into the compute box over USB; I2S firmware hands audio to a XIAO ESP32S3, which forwards it over Wi-Fi or USB; reSpeaker Clip pairs with the phone App over BLE, and the App uploads over Wi-Fi / 4G. The compute box runs whatever must stay local, the speaker plays replies, and the application receives text or voice.
Core Advantages
- An entry box is enough to transcribe
- Add an LLM without changing mics
- Audio can stay on site
- Algorithms tuned per device
| Image | Product | Form | Range | Coverage | Amp | On-board algorithms |
|---|---|---|---|---|---|---|
 | reSpeaker Lite | Desk · dual-mic | ~3m | 180° | 5W | AEC · DoA |
 | reSpeaker XVF3800 | Placed · circular quad-mic | ~5m | 360° | 5W | AEC · DoA · Multi-beam |
 | reSpeaker XVF3800 (with Case) | Placed · circular quad-mic (case) | ~5m | 360° | 5W | AEC · AGC · DoA · VAD · Multi-beam |
 | reSpeaker Clip | Wearable · dual-mic · Bluetooth to phone app | ~3m | 360° | — | Noise reduction (SpeexDSP) |
 | reSpeaker Flex Circular-4 | Split board · circular quad-mic | ~5m | 360° | 10W | AEC · DoA · Multi-beam |
 | reSpeaker Flex Linear-4 | Split board · linear quad-mic | ~5m | 180° | 10W | AEC · DoA · Multi-beam |
XVF3800 in a room, Clip when worn, Flex built into devices
All arrays process audio on the board before it reaches the compute box. Barge-in conversation uses XVF3800 or Flex, whose on-board AEC removes the device's own playback.
Core Advantages
- XVF3800 for placement in a room
- reSpeaker Clip for wearing on the speaker
- Flex split board for building into a robot
- reSpeaker Lite for a desk
| Image | Product | Runs locally | Fits |
|---|---|---|---|
 | reRouter CM4 Series | VAD · ASR (CPU) | Single-site transcription; local relay for cloud conversation |
 | reComputer RK3576 Series | VAD · ASR · TTS (NPU) | Multi-channel transcription (12 channels measured); Chinese/English conversation with a cloud LLM |
 | reComputer RK3588 Series | VAD · ASR · TTS; 4B–7B LLM with an NPU card | Conversation beyond Chinese/English; contact a solution expert for the NPU card |
 | reComputer J30 Series | VAD · ASR · TTS; ~2B LLM | Shares the board with vision or other AI workloads; LLM in the cloud |
 | reComputer J40 Series | VAD · ASR · TTS · 4B–7B LLM | Fully local conversation, offline after setup |
An entry box transcribes; a local LLM needs RK3588 with an NPU card or Jetson
Transcription needs only an entry box. The RK series runs local VAD, ASR and TTS; a local LLM needs RK3588 with an NPU expansion card or a Jetson.
Core Advantages
- Transcription only: an entry box is enough
- RK series runs VAD, ASR and TTS locally, LLM in the cloud
- All local: RK3588 with an NPU expansion card, or Jetson
| Image | Product / output | Purpose | Connection | Route |
|---|---|---|---|---|
 | Mono Enclosed Speaker 4Ω 5W | Plays voice replies | JST speaker port on reSpeaker Lite / XVF3800 / Flex (on-board amp 5W / 5W / 10W) | Conversation, barge-in conversation |
| — | Transcript text | Each final transcript is uploaded to the business system | Ethernet / Wi-Fi | Transcription |
| — | LLM endpoint | Recognized text is sent to a cloud or OpenAI-compatible model API | Network | Speech local, LLM in the cloud |
Conversation needs a speaker; transcription outputs text
The speaker connects to the microphone board, so replies and echo cancellation share one board. Text and model calls leave through the compute box network port.
Core Advantages
- One board handles speaker and echo
- Transcription sends text only
- Cloud LLM never sees the audio
FAQ
Choose capture and transcription for searchable records. Choose conversational voice AI when a device must listen, answer and be interrupted.
Yes. Local transcription and conversational pipelines can run in the deployment environment when supported by the selected hardware and software path.
Start with pickup range, language and interaction requirements, then confirm the host against the reference design and software matrix.
Seeed can evaluate enclosure, branding, packaging and interface changes as a customization project.
Didn't find your answer?
Contact Us
We Are Glad to Be Your Hardware Partner !