For AI conversations
Select Voice DSP+ as the microphone and speaker in your AI application, then talk naturally while the assistant answers aloud.
Sale!
RASPIAUDIO Voice DSP+
Original price was: 99,90€.89,90€Current price is: 89,90€.
Limited pre-order - estimated shipping from 15 September 2026.
LIMITED PRE-ORDER
Estimated shipping starts 15 September 2026.
This item is not available for immediate dispatch. Quantities in the first production run will be limited, and your order reserves a unit. The design is complete and working; in early August 2026 we will set the production quantity from the pre-orders received.
Estimated shipping starts 15 September 2026.
This item is not available for immediate dispatch. Quantities in the first production run will be limited, and your order reserves a unit. The design is complete and working; in early August 2026 we will set the production quantity from the pre-orders received.
A dedicated four-microphone sound card for AI voice, dictation and video calls. Plug it into USB as a driver-free microphone and speaker, or connect it to a Raspberry Pi 40-pin header for direct I2S audio, SPI boot and I2C control.
Voice DSP+ in 90 seconds: USB plug-and-play, Raspberry Pi control, plus real AEC OFF/ON, noise-suppression OFF/ON and 16 m far-field demonstrations.
RASPIAUDIO Voice DSP+
The dedicated sound card for AI voice and video calls.
Four microphones, speaker playback and real-time voice processing on one board. Select it as the microphone and speaker in compatible AI voice and conferencing apps—for example ChatGPT voice conversations, Gemini Live, Claude voice mode, Zoom, Teams or Meet.
Limited pre-order
Reserve a unit now – estimated shipping starts 15 September 2026.
This is a backorder and will not ship immediately. Quantities in the first production run will be limited. The design is complete and working; in early August 2026 we will confirm how many units to manufacture from the pre-orders received.
USB plug-and-play
No product-specific driver
Raspberry Pi 40-pin mode
AEC + beamforming
4-microphone array
More than a microphone array
Voice DSP+ is a complete two-way audio interface made for products that must listen and speak at the same time. It captures your voice, processes it, sends clean audio to the application and plays the response through a speaker or headphones.
For calls and meetings
Use one device for voice capture and playback in Zoom, Google Meet, Microsoft Teams and other conferencing tools.
For embedded products
Build it into a robot, kiosk, tabletop assistant, room controller or any Raspberry Pi voice product.
One voice front end, six product categories
Voice DSP+ combines processed voice capture, audio playback and embedded control, so the same board can move from a USB prototype to an integrated Raspberry Pi product.
Robots
Add listening, speech capture, sound direction and a real speaker output to desktop, mobile, service or companion robots.
Voice assistants
Use the processed microphone channels for speech recognition and the onboard or external speaker path for natural replies and prompts.
Kiosks and smart signage
Build self-service terminals, vending machines and interactive displays with front-facing LINE or room-aware SQUARE microphone geometry.
Conference and intercom
USB microphone and playback, beamforming, AEC and noise reduction support clear two-way conversations from one compact audio board.
Smart toys and companions
Give educational devices, tabletop companions and interactive objects reliable voice input plus built-in audio feedback.
Industrial and retail terminals
Create voice-control front ends for counters, workshops and commercial equipment, with enclosure acoustics validated for the final environment.
Two modes: USB plug-and-play or Raspberry Pi for full control.
1. USB plug-and-play mode
Connect the board to a Windows, macOS or Linux computer. It is detected as a standard USB microphone and speaker, with no product-specific driver to install.
Change or restore the USB firmware from your browser.
Choose a mode, connect Voice DSP+ and install it directly from Chrome or Edge. No command line or developer tools required.
Launch the Web Updater- Recommended first experience: 48 kHz USB Audio Class.
- Two processed microphone channels and stereo playback.
- Voice processing, capture and playback work together out of the box.
- Ideal for AI voice apps, calls, prototyping and product evaluation.
With the current validated firmware, the device may appear as PI AI MIC Assistant Auto.
2. Advanced Raspberry Pi mode
Mount Voice DSP+ on the standard Raspberry Pi 40-pin connector for a direct board-level audio path, without adding a separate USB audio adapter.
- Direct I2S capture and playback.
- SPI firmware boot from the Raspberry Pi.
- I2C host control for diagnostics and product tuning.
- Access to buttons, LEDs, microphone geometry and optional direction feedback.
Start with USB for the fastest result, then move to Raspberry Pi mode when your product needs direct I2S, SPI and I2C access.
One coordinated voice pipeline
Every part of the path is designed to work together, including the playback reference used for acoustic echo cancellation.
4 microphonesCapture speech in LINE or SQUARE geometry.
XVF3800 DSPApplies beamforming, echo cancellation, noise processing and gain control.
USB or I2SSends two processed voice channels to the host application.
Speaker or headphonesPlays the assistant, call or prompt back to the user.
A programmable voice DSP
The XMOS XVF3800 does the hard audio work in real time, before sound reaches the AI or call application.
Acoustic echo cancellation
AEC uses the playback signal as a reference so the microphone path can suppress the sound produced by the device itself.
Beamforming and noise processing
Four microphones work together to focus on speech and improve voice capture in real rooms.
Automatic gain control
AGC keeps speech at a more consistent level as the user moves or speaks more softly.
Automatic microphone geometry
At cold boot, firmware detects the installed LINE or SQUARE module and loads the matching DSP geometry.
SQUARE is designed for voices around the product. LINE is designed for users mainly in front. Change microphone modules only while the board is powered off.
Hear the processing. Test the distance.
Hear real captures switch from unprocessed audio to Voice DSP+ noise suppression and acoustic echo cancellation.
Noise Suppression: before / after
A TV speech source is played beside steady pink noise that simulates a fan. The recording switches to Voice DSP+ noise suppression at 7 seconds.
0:00-0:07Unfiltered: speech + pink-noise background
0:07-0:17.5DSP Noise Suppression enabled
Listen at 7 seconds: the fan-like background falls while the speech remains intelligible. Headphones recommended.
Download the original lossless WAVAcoustic Echo Cancellation: OFF / ON
A hip-hop beat plays through the Voice DSP+ speaker while a separate voice source is captured by the four-microphone array. With AEC enabled, the DSP uses the speaker playback reference to reduce that same beat in the microphone signal while preserving the voice.
0:00-0:05.21AEC OFF
0:05.21-0:12.43AEC ON
0:12.43-0:19.64AEC OFF
0:19.64-0:21.00AEC ON
0:21.00-0:26.85AEC OFF
The first two seconds of the synchronized recording have been removed, leaving a 26.85-second comparison. The final AEC ON section ends at exactly 21 seconds in the video, then AEC OFF remains active through the end. The speaker beat falls during each AEC ON section while the external voice remains. This enables wake words, voice commands and dictation while music, prompts or responses are playing. Headphones recommended.
Hey Jarvis far-field wake-word demo
The same wake phrase is repeated while the speaker moves progressively farther from Voice DSP+, from 1 m (3.3 ft) to 16 m (52.5 ft). The split screen keeps the test distance visible on the left and the Voice DSP+ hardware fixed on the right. Watch the indicated board LED change colour when the wake word is detected.
1 m / 3.3 ft4 m / 13.1 ft7 m / 23.0 ft10 m / 32.8 ft13 m / 42.7 ft16 m / 52.5 ft
How this demo works: the pretrained hey_jarvis model runs on the Raspberry Pi, not in the XMOS firmware. The XVF3800 provides beamforming and the processed 16 kHz voice front end; the Raspberry Pi performs wake-word inference. No additional noise-suppression stage is added on the Pi. View Raspberry Pi setup, architecture and source.
Test conditions: The Noise Suppression demo uses TV speech with a steady pink-noise, fan-like background. The AEC comparison uses two separate 30-second, level-matched microphone captures of a voice source while a hip-hop beat plays through the board speaker. The far-field demo is an outdoor wake-word test at marked positions from 1 m (3.3 ft) to 16 m (52.5 ft).
Capture, process and play voice on one board
Voice DSP+ is a complete voice interface, not just a microphone array. It can work as a standalone USB sound card or become the audio front end of a Raspberry Pi product.
The main board combines capture, real-time voice processing and audio playback. The two microphone boards can be detached and repositioned.
- Four digital microphonesTwo detachable boards carry two microphones each. Arrange all four microphones in LINE or SQUARE geometry to suit the product.
- XMOS audio DSPReal-time beamforming, acoustic echo cancellation (AEC), noise processing, automatic gain control (AGC) and direction of arrival (DoA).
- USB-C microphone and speakerConnect Voice DSP+ directly to Windows, macOS or Linux and select it as a standard microphone and playback device. No product-specific driver is required.
- Onboard speaker included – standalone from day oneThe built-in speaker is ready for voice prompts, testing and compact products without any extra audio hardware. Its output is intentionally limited to roughly the experience of a good smartphone: clear and practical for voice, but not intended for powerful room-filling sound.
- Up to 11 W output for an external passive speakerFor louder and fuller sound, disable the onboard speaker and connect a suitable 4 Ω or 8 Ω passive speaker to the onboard amplifier output. Available power depends on the supply voltage, speaker impedance and speaker rating.
- Stereo headphone outputThe 3.5 mm jack is automatically detected; the validated firmware routes audio to the headphones and mutes the speaker amplifier.
- Optional Raspberry Pi 40-pin connectionUse direct I2S audio, SPI firmware boot and I2C control when deeper embedded integration is required.
- Programmable user buttonReady for push-to-talk, dictation, wake-up or application-specific interaction.
The complete Voice DSP+ core kit
The main board, all four microphones and the slim enclosure are included. Add a USB-C cable and you are ready to start in USB mode.
Voice DSP+ main board
The complete sound card with XMOS voice DSP, USB and Raspberry Pi interfaces, onboard speaker, stereo headphone output and external speaker amplifier.
Two-microphone boards
Two detachable microphone bars, each carrying two digital microphones: four microphones in total, ready for LINE or SQUARE geometry.
Slim 3D-printed case
A compact enclosure designed for a clean, protected and ready-to-demonstrate USB build.
Included: Voice DSP+ main board, two detachable two-microphone boards and slim 3D-printed case. Raspberry Pi, USB cable, external speaker, LED ring, laser sensor and extension components are not included unless explicitly listed for another bundle.
Technical overview
| Voice processor | XMOS XVF3800 programmable voice DSP. |
|---|---|
| Microphones | Four digital microphones on two detachable two-microphone modules. |
| Microphone geometry | LINE or SQUARE; automatically detected and configured at cold boot. |
| Voice processing | Beamforming, acoustic echo cancellation (AEC), noise processing, automatic gain control (AGC) and direction of arrival (DoA). |
| USB audio | USB Audio Class, 48 kHz recommended firmware, two processed microphone channels and stereo playback. No product-specific driver required on current Windows, macOS or Linux systems. |
| Raspberry Pi audio | Direct I2S through the 40-pin header; recommended firmware uses 16 kHz with two processed capture channels and two playback channels. An advanced 48 kHz profile is also published. |
| Host interfaces | I2S audio, SPI firmware boot and I2C host control at address 0x2c in Raspberry Pi mode. |
| Audio outputs | Mono differential speaker path and stereo headphone output. Inserting a headphone jack automatically mutes the speaker amplifier in the validated firmware. |
| User controls | Onboard user button; the recommended USB firmware sends Ctrl+Shift+D while the button is held. |
| Visual feedback | Status LEDs plus support for an optional 24-pixel WS2812B direction-of-arrival ring. |
| Firmware resources | Validated USB and Raspberry Pi firmware files, checksums and setup instructions are published on GitHub. |
Board, connectors and operating modes
Code and documentation on GitHub
The public repository includes installation scripts, validated firmware, hardware documentation, pinout and integration diagrams, checksums, Raspberry Pi integration and the board-specific source overlay.
Check pinout, firmware and integration on GitHubThe essentials, at a glance
Do I need a Raspberry Pi?
No. USB mode works directly with a current Windows, macOS or Linux computer. A Raspberry Pi is only needed for the advanced embedded mode.
Do I need to install a driver?
No product-specific driver is required on current Windows, macOS or Linux systems. Select the board as both the microphone and speaker in your application.
Can I change the microphone geometry while it is powered?
No. Power the board off before changing the microphone modules. The selected geometry is detected only at cold boot.
Is the complete firmware fully open source?
The public repository includes RASPIAUDIO’s board-specific source overlay, firmware, installer and documentation. The complete XMOS SDK is distributed separately under XMOS licensing.
Start with USB. Go deeper with Raspberry Pi.
Get a premium voice experience out of the box, then keep the same hardware when your prototype becomes a fully integrated product.
Build a gesture-controlled ChatGPT dictation assistant
A focused voice terminal
Add a more powerful speaker and a laser time-of-flight range sensor to create a dedicated voice assistant. In this example, passing a hand above the sensor triggers the configured ChatGPT dictation control, then Voice DSP+ captures processed speech for transcription.
- Hand detected above the device.
- The Raspberry Pi triggers the configured dictation control.
- Voice DSP+ captures and processes speech.
- The transcript is sent to the ChatGPT application.
The higher-power speaker and laser range sensor are extension components for this example build.
Video coming soon
Gesture-controlled voice assistant demo
This video will show USB setup, hand detection, ChatGPT dictation activation and the complete listen-and-reply experience.
Related products
-
-
-
Sale!
ESPMUSE programmable speakersMuse Radio
Original price was: 99,00€.89,00€Current price is: 89,00€. Add to cart -
