Speech enhancement for non-stationary noise around a machine cabin
Ladda ner
Typ
Examensarbete för masterexamen
Master's Thesis
Master's Thesis
Program
Sound and vibration (MPSOV), MSc
Publicerad
Författare
Zhou, Yiliang
Modellbyggare
Tidskriftstitel
ISSN
Volymtitel
Utgivare
Sammanfattning
This thesis is mainly concerned with the solution for speech enhancement in the
presence of non-stationary noise around the machine cabin. This allows outside
speech to enter the cabin and reduces unwanted noise. The application scenario
of this work is a signal processing system using a microphone array outside the
cabin to capture signals and using different algorithms to enhance the speech signals.
With the help of this system, the machine operator is able to get speech
information in a noisy environment. The noise sources in this situation are more
complex and non-stationary. Examples of noise sources include engine noise, traffic
noise, or other construction activities. Previous work done by Tomoya chose the
microphone array configuration and developed the beamforming method. From the
results of the valuable work, it is found that beamforming is able to increase the
signal-to-noiseratio (SNR) in the current situation, but the sound quality and SNR
are still limited due to the low input SNR and non-stationary noise environment.
Therefore, modified beamforming and new methods are implemented in this work.
noise cancellation(NC) predicts the transfer path for the noise signal and removes
it by controlling the minimum error of the output. Noise suppression(NS) uses the
scheme of spectral subtraction to subtract the noise spectrum from noisy speech
spectrum. A combination of beamforming and noise cancellation and a combination
of beamforming and noise suppression method are developed and evaluated. The
result shows a better performance for this low input SNR and non-stationary noise
case.
Beskrivning
Ämne/nyckelord
Speech enhancement, Non-stationary noise environment, Beamforming, Noise cancellation, Noise estimation