Ultra-Low Power Neuromorphic Front-End Chip for Voice Processing
NeuroVoice, a miniature Neuromorphic Analog Signal Processor, addresses the most annoying problems of mass-market hearables: hearing quality in the presence of irregular noise, power consumption, cloud connectivity, and data privacy issues.
NeuroVoice Product Family
Voice Activity Detection
Most state-of-the-art voice detection devices are based on digital signal processing and consume a lot of power. Voice activation solutions are inefficient in noisy environments.
The NASP NeuroVoice VAD is an ultra-low-power neuromorphic chip that detects voice in any noisy background. It is always-on, and no internet is needed. Applications include earbuds, wearables, voice access, IoT, Industry 4.0, robotics, Smart Home/Factory, mobility, and more.
| Parameter | Value |
|---|---|
|
Power consumption, µW |
~35 |
|
Size, mm2 |
1.2 |
|
Input/Output |
PDM/I2S/GPIO (voice flag)/PDM/I2S |
| Feature | Description |
|---|---|
|
Voice Passthrough |
Passes voice and mutes background sounds |
|
Background Signal Bypass |
WWD command to activate transparent voice bypass |
|
Speech/Command Intelligibility in Noisy Space |
Control Word in noisy environment |
Smart voice control
State-of-the-art voice control systems rely on cloud connectivity to be efficient in noisy environments. However, this is not energy-efficient and is disliked by users.
The brain-inspired NASP NeuroVoice Smart Voice Control is an ultra-low-power chip that continuously processes the microphone signal and detects voice commands. Even in high-noise environments, the chip processes all data on-device. There is no need for a connection to the cloud.
| Feature | Description |
|---|---|
|
Voice Passthrough |
When voice is detected all other background sounds are muted. Perfect for voice command control and clear communication |
|
Background Signal Bypass |
WWD command to activate transparent voice bypass |
|
Speech/Command Intelligibility in Noisy Space |
Increases voice command intelligibility for WWD/KWS functionality in noisy environments. Solves the problem of command recognition in noisy signals for Smart Home, Smart Factory, Wearables and more |
Speaker Recognition
Voice-aware identity authentication is advantageous for various devices, from personalized home appliances, parental content controls, health trackers, to two-way radios for group recreation activities.
On-device voice processing is becoming more viable, especially for privacy-sensitive applications. Speaker recognition performed with minimal power consumption is critical for always-on functionality in mobile and wearable contexts.
The NASP NeuroVoice Speaker Recognition is an ultra-low-power neural core that continuously processes the microphone signal and extracts voice patterns to be compared with samples of known voices. Even in noise environments, all data is processed on-device, with remarkably low false positives.
| Feature | Description |
|---|---|
|
Voice Authentification |
A voice pattern is extracted and compared with previously learned known voice samples |
|
Recognition Clarity |
Combined with command detection, it practically eliminates false positives. Perfect for voice management and clear communication |
|
Ultra-Low Power Consumption |
Always-on voice recognition systems are usually power-hungry. The NeuroVoice’s ability to perform it at ultra-low power consumption is unique |
Voice Extraction
Voice Extraction provides the best hearing experience! Always-on, adaptive to different types of noise, ultra-low-power, and on-device, cloud-independent voice processing
The NASP Voice Extraction core, powered by neural networks, extracts human voice from any signal, enabling clear communication between speakers and listeners. It is distinct from noise cancellation and focuses on real-time voice extraction by neural networks without the need for environmental microphones or DSP systems.
| Feature | Description |
|---|---|
|
Voice Passthrough |
When voice is detected, the operation mode is changed according to the configurable preset list. Perfect for intelligent hearing aids and TWS with voice enhancement. |
|
Background Signal Bypass |
In absence of voice, this feature enables the background sounds to be passed through, which is important for hearing aids use case. |
|
Speech/Command Intelligibility in Noisy Space |
Increases speech intelligibility for hearing support. Solves the problem of low intelligibility in noisy spaces for many people. |
Hearing difficulties are common today. Mild to moderate hearing problems often coincide with challenges in hearing in noisy environments.
Refining the ability to hear the human voice in noisy environments would be a significant step forward for hearing aids, hearing assistance, conferencing, two-way radios, or even gaming applications.
Voice extraction by a neural network is the answer if it is implemented not in the cloud but on the device. NeuroVoice, a small low-power and low-latency chip, offers precisely that.
Digital hearing solutions have challenges:
1Digital sound processing is power hungry
2Digital solutions can’t provide low latency needed for real-time voice processing
3Active noise cancellation is helpless against irregular noises
1Ultra-low power consumption below 150µW
2NASP ensures ultra-fast inference in up to 20µs
3Voice extraction delivers clear voice in any background conditions
Digital Signal Processing vs NASP Capabilities
| Solution | Acoustic Voice Processing | Mixed Voice Processing | Irregular Noise | Standard Noise | Single Microphone Solution | Transparent Mode AI Support |
|---|---|---|---|---|---|---|
|
DSP |
Yes |
No |
No |
Yes |
No |
No |
|
NASP |
Yes |
Yes |
Yes |
Yes |
Yes |
Yes |