Intel Host-Based Microphone Array

• Goal
  - To integrate a host-based low-cost, high performance microphone array into platform to better support  conferencing and speech recognition applications
• Key Results
  - Azalia audio architecture definition for ICH6-ICH8 (joint work w/DPG)
  - Florence Microphone Array for next generation Mobile PC (joint work w/MPG)
  - Microphone Array algorithms research (AEC, Source Tracking, Beamforming, Post-Filtering, BSS)
  - Production microphone array planned in 2005

On-Line Comparison of Microphone Arrays for Speech Recognition Applications

• Can Florence microphone array improve 3rd party speech recognition accuracy in noisy environments?
• Compared with best industry microphone array products, how is about Florence microphone array?

Summary

• Compared with best industry microphone array products, Florence microphone array consistently demonstrates equal or better performance in both noise reduction and speech recognition.
• Compared with laptop microphone, Array Microphone provides better audio
• Florence microphone array is significantly cheaper than industry microphone array products and provides hands-free capability

Real-Time System for Blind Source Separation (BSS)

• BSS: Advanced Microphone Array Processing Algorithm
  - What’s BSS?
  - - To extract specified source signal from the mixtures of source signals only by exploring the statistical independence of source signals in mixed array microphone signals
  - BSS vs. Beamforming (BF)
  - - Don’t need any information about the array geometry
  - - Be able to separate the source signals coming from same direction
• Real-Time BSS system/demo:
  - Hot but very challenging research topic! By now very few research groups can demonstrate real-time BSS systems in the world.
  - We have built the best real-time BSS system, and can demonstrate
  - - Completely separate the source signals as BF does (better than BF)
  - - Be able to separate sources coming from same direction;
  - - 2 microphones to separate only 2 acoustic sources

Summary

• We have demonstrated the best real-time BSS system, which could
  - Completely separate two speakers’ voice
  - Even can separate two sources in one direction
  - Do not need priori knowledge about sources, channel and array geometry
• Demonstrated our leadership not only at practical algorithm research (BF, AEC) but also at leading-edge research front line (BSS)