Clear Audio

Clear in noise, and you can interrupt while it speaks

Local Compute

Fully local: 0.9 s to reply, measured; works offline

Tuned, Ready to Run

Tuned and benchmarked per board, open source

Scenario details

Far-field pickup with barge-in, on-device recognition, voice-driven equipment.

Pickup & Barge-in

Pain Points & Challenges

When kiosk speakers playback audio, raw microphones easily suffer severe acoustic feedback from their own output, blinding the device to user speech and forcing visitors to wait until lengthy prompts finish before speaking again.

Edge AI Solution

Integrate hardware-grade AEC array boards that cancel the local playback reference directly at the chip level, capturing user barge-in utterances instantly even during 80dB loud speaker output to cut playback immediately.


Core Advantages

  • Hardware AEC: Cancels onboard speaker playback up to 80 dB in real time
  • Sound DoA: Computes arrival angles to steer beamforming toward talkers
  • Instant Barge-In: Full-duplex engine cuts playback in milliseconds on user speech

Hardware AEC

Sound the equipment plays itself is removed on the array board, so it is never taken as a command. Motor and gear noise is not removed.

DoA & Beamforming

The board outputs direction of arrival (DoA) and applies beamforming, with pickup up to about 5 m; circular arrays cover 360°, linear 180°.

Barge-in

The microphone keeps listening during the reply; when the user speaks, playback stops and the next turn starts. Barge-in can be turned off in settings.

Interruptible Conversational Voice AI

Ready to deploy
$npx -y @seeed-studio/solution-cli deploy conversational_voice_ai
View full design & BOM →

On-Device Recognition

Pain Points & Challenges

Public cloud LLM backends crash completely during internet outages or within isolated private networks, per-token API billing inflates long-term operating expenses, and enterprise sensitive inquiries risk data-compliance violations.

Edge AI Solution

Deploy edge hosts with on-board NPUs running offline ASR, quantized Small Language Models (SLMs), and local TTS entirely on-premise, eliminating cloud reliance and locking round-trip response latencies under 1 second.


Core Advantages

  • Multilingual: Supports 30 dialogue languages with sub-second on-device latency
  • Offline Privacy: On-device LLM processes conversations with zero internet reliance
  • Custom Voices: Offers tiered synthesized voices with optional rapid voice cloning

Multilingual Recognition

30 conversation languages, one chosen at deployment; if the device cannot serve the chosen language, deployment stops with an error.

On-Device LLM

The LLM runs on the local host; conversation text stays on the device.

Voice Persona

Three voice tiers, robotic, synthetic and natural; the natural-voice model can clone a voice from reference audio.

Conversational Desktop Robot

Ready to deploy
$npx -y @seeed-studio/solution-cli deploy reachy_claw_voice_robot
View full design & BOM →

Voice-Driven Equipment

Pain Points & Challenges

Embodied robotic arms and service robots relying on touchscreen inputs fail to free operators' hands, while cloud-assisted voice control suffers variable network jitter that leads to delayed grasping motions and physical collisions.

Edge AI Solution

Embed compact edge compute modules directly inside robots, parsing spoken commands into structured coordinate frames and trajectory codes on-device, dispatching motion controls over internal high-speed buses in tens of milliseconds.


Core Advantages

  • Name-and-Grasp: Verbal item names dispatch directly to robot arms over fast bus
  • Zero-Retrain: Add new objects via plain config without retraining models
  • Fuzzy Intent: Semantic mapping aligns colloquial variations into exact device commands

Grasp by Name

About 11 s from voice command to object returned; grasps boxes, opaque cups, standing opaque bottles, bananas and oranges.

No Retraining

The detector's object list is a config item, so adding an object is a config edit; transparent objects and objects wider than 10 cm are out of range.

Devices and Systems

Extensible commands; loose wording maps to the closest one, then goes to the device or system.

Deployment & Selection

The configurator works through use case, voice tier, concurrency, on-device LLM, microphone setup and latency target to reach specific models.

Not sure which to choose? See the full selection guide →

Use Case

Configure Your Solution

Step by step, narrowing down to your recommended BOM

  • Wake-Word Frontend Kit

    Wake/command words only; pair w/ your own host

    硬件/物料名称数量型号/SKU说明
    XIAO ESP32-S3 Sense (Built-in Microphone)1113991115-
  • Local Transcribe Kit

    Single-lang local STT, no TTS. For meeting notes & voice memos.

    硬件/物料名称数量型号/SKU说明
    ReSpeaker Lite (2-Mic Near-Field)1107990273-
    reRouter CM4 1020321110110205-
  • Two-Way Dialogue Kit

    Multi-lang ASR + TTS; cloud or local LLM optional.

    硬件/物料名称数量型号/SKU说明
    ReSpeaker Lite (Near-Field ≤3 m)1107990273-
    reRouter CM4 102032 — Machine Voice Entry (Single-lang)1110110205-

Architecture

Choose front-end, hybrid, or all-in-one

Voice compute placement determines capability ceiling & per-unit BOM. Three common models:


Core Advantages

  • Front-End Only: Lightweight modules handle wake word and AEC for your host
  • Hybrid Setup: Edge handles responsive audio while cloud models run reasoning
  • All-in-One: On-device compute powers full pipeline for strict privacy needs
ImageHardwareVoice Concurrency (voice-only)Voice CapabilitiesLocal LLM
reRouter CM4 SeriesreRouter CM4 Series1 channelMachine VoiceCloud optional
reComputer RK3576 SeriesreComputer RK3576 Series1 channelSimulated Voice4B–7B (w/ RK1828 card)
reComputer RK3588 SeriesreComputer RK3588 Series1 channelSimulated Voice4B–7B (w/ RK1828 card)
reComputer J30 SeriesreComputer J30 Series1 channelNatural Voice~2B
reComputer J40 SeriesreComputer J40 Series2–3 channelsNatural Voice4B–7B
reComputer J50 SeriesreComputer J50 Series3+ channelsNatural Voice7B–14B
NVIDIA Jetson AGX Thor Developer KitNVIDIA Jetson AGX Thor Developer Kit6+ concurrent (voice-only)Realistic voice14B–32B

Local Voice Hosts

Scale voice hosts from CM4 to AGX Thor

The table selects the device for the Voice Host (ASR / TTS / Agent) block in the middle of the architecture diagram, listed by family: voice-only concurrency, voice capability, and local-LLM size per family. Exact models, configurations and accessories are picked in the configurator below (mic & speaker selection: next tab). Note: running a local LLM beside the voice pipeline reduces performance and adds latency; for high concurrency plus large models, choose a higher compute tier.


Core Advantages

  • Entry Tiers: CM4 and RK3576 run single-stream wake words and offline voice
  • Professional Jetson: J30 and J40 enable realistic TTS and on-device LLMs
  • Flagship Thor: AGX Thor handles multi-channel voice and heavy-parameter models
ImageProductTypeBest forKey specs
ReSpeaker LiteReSpeaker LitePickup (near-field)≤ 3 m / desktop / single workstation2-mic array / on-board AI audio processing / USB · I²S
ReSpeaker XVF3800ReSpeaker XVF3800Pickup (mid/far-field)3–5 m / meeting room / living room / workstation4-mic / XMOS DSP / AEC / ESP32S3 wake-word front end
reSpeaker ClipreSpeaker ClipPickup (wearable)Worn on the body / up to 3 m / 360°2-mic PDM / SpeexDSP noise suppression / Bluetooth LE 5.3 + Wi-Fi
ReSpeaker Flex Circular-4ReSpeaker Flex Circular-4Pickup + playback (circular)Built into devices, 360° / wake-word front end included4-mic / XMOS DSP / AEC / ESP32S3 / 10 W amplifier
ReSpeaker Flex Linear-4ReSpeaker Flex Linear-4Pickup + playback (linear)Built into devices, 180° / wake-word front end included4-mic / XMOS DSP / AEC / ESP32S3 / 10 W amplifier

Mics & Speakers

Match mic and speaker by range and form

The key variables for microphones are pickup distance and ambient noise. Key specs and recommended pickup + playback combinations are below.


Core Advantages

  • Diverse Formats: Offers desktop, embedded split-board, and body-worn arrays
  • On-Chip Audio: Hardware DSP integrates echo cancellation, noise filter and DoA
  • Open Integration: Standard interfaces and open-source firmware accelerate builds

FAQ

Didn't find your answer?
Contact Us
We Are Glad to Be Your Hardware Partner !
Next