ClearVoice vastly improves automatic speech recognition (ASR) for voice command and control user interfaces (UI), and artificial intelligence (AI) driven natural language processing (NLP) found in services such as Amazon Alexa Voice Serices (AVS), Microsoft Cortana and Google Assistant. ClearVoice is an advanced software solution for voice IoT applications that enables superior voice recognition in far-field and noisy environments via patented voice signal processing algorithms for multi-microphone adaptive beamforming, noise cancellation (NC), acoustic echo cancellation (AEC) and automatic gain control (AGC). ClearVoice offers broad hardware platform support and highly optimized CPU and memory utilization for easy system integration with reduced cost and complexity. It supports single microphone and arrays from 2 up to 8 or more microphones.
Features
- Multi-platform Support: MIPS/ARM/x86
- Multi-OS support: Android/iOS/Windows/Mac/Linux
- WebRTC support: Integrated in WebRTC framework
- APIs: well-defined APIs for integration with 3rd party applications/products
Benefits
- Software only solution, eliminating the need for an external chip
- Low CPU usage, optimized for MIPS/ARM/x86 platforms
Applications
- NC for IoT (Internet of things), mobile devices and automotive applications
- AEC for various PC and embedded platforms - Windows, Mac, Android, iOS - and embedded full-duplex voice communication systems
- Line Echo Cancellation (LEC) for CPE (Consumer Premises Equipment) - ATA , set top box, residential gateway
Specifications
- Acoustic Echo Cancellation (AEC)
- Compliant with ITU-T G.167, P.340, 3GPP TS.26.131, TS.26.132, and ITU-T G.160 standards
- Cancels echo spans of up to 128ms in a 512ms moving window
- Greater than 90dB echo cancellation during single talk
- Very low (6ms) algorithmic delay
- Supports narrowband, wideband (7 KHz) and super wideband (14KHz) voice
- Auto-tune based on platform audio interface characteristics - for commercial off-the-shelf platforms
- Very fast (16 samples) double talk detection
- Noise Cancellation (NC)
- Noise suppression >30dB with minimal distortion
- Convergence time is less than 300ms
- Provides high SNR without musical noise
- Automatic Gain Control (AGC)
- ITU-T G.169 compliance
- Amplifies or attenuates only speech segments
- Programmable target power level in the range 0 to -30 dBm0
- Multi-MIC Adaptive Beamformer
- Supports 2 or more microphones
- Supports single (low complexity) and multiple source tracking
- Forms very sharp beam to preserve the speech intelligibility even under high noise
- Improves far-field speech recognition accuracy rate significantly
- Employs side lobe estimation to preserve intelligibility for more than 2 microphones
- Robust against small errors in the direction of arrival (DOA) estimation
- Significant performance improvement is observed with
third party voice services for noisy and far-field signals
- Amazon's Alexa Voice Service (AVS)