| Summary: | 碩士 === 臺灣大學 === 電機工程學研究所 === 95 === We use a flight recorder to assist the aircraft accident investigation for identifying the probable cause of aircraft accident. There are two forms in a flight recorder, the Flight Data Recorder (FDR) and the Cockpit Voice Recorder (CVR). The CVR effectively acts as a latent signal transducer for both speech and non-speech acoustic information such as propeller noise, switch sound…etc. When the power of the noise from propeller is overwhelmingly greater than the power of the desired pilot’s speech, we cannot identify the pilot’s speech content to investigate the accident cause. A typical technique, adaptive noise cancellation, appears to be the primary task, but it assumes the use of two sensors. In our situation, we do not have the second sensor to pick up the reference signal to achieve noise cancellation. This thesis hence focus on single channel noise cancellation using adaptive filter based on principal frequencies synthesis (PFS) and linear prediction error filter (LPEF). Adaptive filter based on PFS is used to cancel the propeller noise form CVR; LPEF is used to whiten the residual noise in speech. Thus we can identify the pilot’s speech content more easily from the enhanced speech to assist us in the accident investigation.
|
|---|
