Speech Enhancement Under Broadband Noise

Speech enhancement is an important task for many signal processing applications,including speech recognition,mobile communications and hearing aids.Compared with narrowband noise,broadband noise has a greater impact on the quality and intelligibility of speech.Especially under the condition of low signal-to-noise ratio,broadband noise can cause serious interference and destruction to pure speech.This paper proposes an enhancement method for single-channel speech signals based on harmonics extraction.Given the fact that most of the energy of speech signal is concentrating on harmonic structures of voiced frames and the quality and intelligibility of speech depend on the integrity of harmonic structure in speech to a large degree,we put forward a strategy to precisely locate the harmonic structures of voiced frame in frequency domain and the harmonic structure obtained is used to improve the IMCRA noise estimation method.And then the modified noise estimation method is utilized to obtain a priori SNR which is used for spectral gain calculation of the spectral gain function we designed.It can retrieve the harmonic structures of voiced speech and suppressing the noise component in the frequency domain.The performance of the proposed speech enhancement method is evaluated by several quality tests with four types of noise.According to the experimental results,the proposed enhancement algorithm achieves better performance than the other three existing speech enhancement algorithms as a whole,particularly at low signal-noise-ratio(SNR)and broadband noise conditions.