POLYN Technology’s Neuromorphic Analog Signal Processor (NASP) NeuroVoice VAD is an always-on, ultra-low-power chip that detects voice in any noisy background, at microwatt-level power consumption and microsecond-scale latency.
Everything happens on the chip, so no Internet is needed. Potential applications include smart remote controls, earbuds, wearables, voice access, IoT, Industry 4.0, robotics, Smart Home/Factory, mobility, and more.
NASP NeuroVoice VAD chip (NV-VAD 100) specifications:
- Features
- Voice detection at ultra-low power consumption
- Voice passthrough – Passes voice and mutes background sounds
- Background signal bypass – Wake Word Detection (WWD) command to activate transparent voice bypass
- Speech/command intelligibility in noisy spaces – Increases voice command intelligibility for WWD/KWS (Keyword Spotting) functionality in noisy environments for Smart Home, Smart Factory, Wearables, etc.
- Audio
- Interfaces – PDM/I2S
- Voice delay detection – 25 ms
- Host interfaces – SPI/I2C used for initial configuration and status readout, VAD pin
- Debugging – Boundary Scan and Debug unit access to core registers via JTAG
- Supply Voltages
- Digital core – 1.2V
- Analog core – 1.2V
- I/O – 1.8V
- Power consumption – 30~35 µW (I assume when active?)
- Power modes
- Standby – NeuroVoice VAD is at full stop, keeping/waiting 
 for configuration; all clocks are stopped, the NeuroVoice VAD is ready 
 to accept commands from the external serial interface (SPI/I2C)
- Sleep – Substate of Standby; keeps the MEMS microphone in sleep mode with the possibility 
of rapid wake-up.
- Low Power – Uses in cases when the MEMES microphone should work, but the VAD Neurocore functionality is disabled
- Active – Fully operational mode; all features are active
- Dimensions – 1.2 mm2
The VAD core operates with 8 msec frames, and the gap between frames can be increased up to 8 frames to further reduce the power consumption. Another configurable parameter is the VAD Sensitivity Level (VSL), which can be set between 0 and 1. This tunes the neural network core to a different balance of voice detection between false positives and false negatives.
There’s limited information about software support and devkits, but the company plans to showcase the solution at CES 2026. and provide information on NeuroVoice evaluation kits and early-access programs for companies developing ultra-low-power voice-controlled products. A few more details may be found on the product page and the press release.
Thanks to TLS for the tip.
Jean-Luc started CNX Software in 2010 as a part-time endeavor, before quitting his job as a software engineering manager, and starting to write daily news, and reviews full time later in 2011.
Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress. We also use affiliate links in articles to earn commissions if you make a purchase after clicking on those links.
