ABOUT PARROT
Parrot Algorithms
Since 1994 Parrot engineers have built state-of-the-art acoustic software to offer superior sound quality in the most demanding of settings, with a focus on high-noise car environments.
Noise Reduction
Characteristics
- Treatment on spectral domain (128 frequency bands).
- Uses statistical properties of speech.
- Continuous noise estimation.
- Well-coupled with our AEC.
Performance
- No musical noise (colourless residual noise).
- Low speech quality degradation.
- Robust to non stationary noise.
- Noise level reduced by more than 10 dB (up to 20dB).
- Initial convergence masked by AGC.
- Fast convergence (~60ms).
Parrot Beamforming®
- The newest Parrot hands free car kits incorporate the very latest “Parrot Beamforming®” signal processing technology by means of a pair of twin microphones. As a result, the kit picks up the driver's voice more clearly, while effectively reducing surrounding noise.
Reliable Voice Recognition
Functions
- Isolated words (name dialing).
- Keyword Spotting (full vocal interface).
- Up to 500 voice patterns.
- Rejection of « out of vocabulary » names.
- No language dependency.
Main Characteristics
- Improved MFCC-based feature extraction.
- DTW matching.
- Noise robustness.
- Noise adaptability.
- Speech/non-speech detection.
Performance
- High recognition rates.
- Algorithms optimized with home-made database, specially difficult for ASR.
Design on DSP
- Hand-written assembler code.
- DTW hardware accelerator (7 times faster).
- Real-time processing (negligible response time after a recognition).
- Noise adaptability.
- Available on TI and Oak DSPs.
Acoustic Echo Cancellation
Characteristics
- Two layers: Variable Step Size LMS & Automatic Gain Control.
- Acoustic filter length : 256 samples (32 ms).
- Low complexity : no double talk detection, only one mode.
Performance
- Mean ERLE : 40 dB (up to 50 dB).
- Robust to double-talk : Full-duplex canceller.
- Fast convergence (31 dB/s).
- Noise adaptability.
- Robust to noise and acoustic condition variations.
|
