|
- Data recorded from n microphones is processed to produce k output signals (often k=1).
- The processing uses phase information to obtain a 3D recording.
Technology
Beam Forming (directional recording):
- Goals:
- Narrow beam.
- Small side lobes.
- Constant beam width over wide frequency range.
- Features:
- Needs large microphone spread for narrow beam in low frequency.
- Needs many microphones (say 8).
- Low complexity.
- Technology advantage:
- Special optimization technique using pre-filtering (presented in ICASSP 97).
Adaptive Noise Cancellation:
- Goal:
- Amplify speaker signal, attenuate noise.
- Features:
- Uses few microphones (n microphones for n-1 noise sources).
- Works with multiple echoes (inside a reverberating room).
- High complexity.
- Needs speaker activity detector.
- Technology advantage:
- Efficient methods for large filters (patent).
- Stable algorithm for LMS methods (patent submitted).
Applications
- Directional recording and spatial effects (speaker tracking, stereo recording, etc.).
- Noise cancellation for hands-free applications (mobile telephony, echo cancellation, front-end for speech recognition engine, etc.).
- Audio conferencing (use PC to replace dedicated expensive conferencing hardware).
Solution Components
- Recording hardware: n microphones with amplifiers and A/D converters.
Special solution for speech microphone array (patent): 8 analog signals can be filtered into 11 KHz each and multiplexed into one 44 KHz stereo signal. This signal can be sampled by a standard PC sound card.
- Processing software that contains:
- Directional recording with manual / automatic control of the preferred direction.
- Adaptive filtering for reducing noise sources.
The software may be implemented in the host PC or on a DSP to be used by any appliance.
Demo
The following are WAV audio file examples, sampled at 8 KHz.
Speaker with radio background noise (313 Kbyte each):
Speaker with another background speaker (231 Kbyte each):
| |
|
