Hybrid Griffiths-Jim ANC 구조를 향상시키기 위한 Two-Paired 3채널 구성 Two-Paired 3-channel Scheme for Improved Hybrid Griffiths-Jim and ANC Enhancement Structure원문보기
Interest on the speech recognition in car environment is rising rapidly due to the prospect of realizing Telematics services using speech interface. However, in order to render useful services, effective speech enhancement must be made for improved speech recognition in noisy car environment. This p...
Interest on the speech recognition in car environment is rising rapidly due to the prospect of realizing Telematics services using speech interface. However, in order to render useful services, effective speech enhancement must be made for improved speech recognition in noisy car environment. This paper presents a 2 paired 3-channel speech enhancement method to improve speech recognition in car noisy environments. Noise reduction techniques for speech enhancement have been widely investigated for various application scenarios and noise conditions. Most of these techniques are based on the assumption that noise and speech are uncorrelated. However, in the real world, noise might not always be uncorrelated with speech signal, especially if reverberations exist due to the geometry of the space. The conventional noise reduction methods involving single and dual channels experience insignificant performance improvement due to this unrealistic assumption. Moreover, because of the varying degree of noise being introduced between speaker and microphone, the performance becomes sensitive to the location and directivity of the channel with respect to speaker. The proposed method can be divided into three modules, cross-correlation, channel-selection, and Griffiths-Jim and ANC combination modules, respectively. First, in the cross-correlation module, the input signal becomes output signal with reduced channel noise, particularly burst noise, using cross-correlation function. Second, the primary and reference pair input signal to the Griffiths-Jim and ANC combination module takes in speech dominant signal as primary input and noise dominant input as reference out of the channel-selective parameter. Finally, the aim of the last module is to reduce the background noise using ANC. The enhanced signal produced by the proposed channel selection method gives better performance independent of between channel distance or distance between speaker and microphone.. For example, the conventional dual channel performance evaluation of Griffiths-Jim and ANC hybrid structure shows that the performance is better with the channels being close to each other. In particular, we propose a channel section scheme that establishes primary and reference channel from the initial 2 paired 3-channel inputs. This step essentially establishes the primary channel being speech dominant while the reference channel becomes noise dominant. The established primary and reference pair become the input to the Griffiths-Jim and adaptive noise cancellation (ANC) combination speech enhancement module. The representative experiments show that the proposed scheme is effective in achieving improved recognition in noisy car environment.
Interest on the speech recognition in car environment is rising rapidly due to the prospect of realizing Telematics services using speech interface. However, in order to render useful services, effective speech enhancement must be made for improved speech recognition in noisy car environment. This paper presents a 2 paired 3-channel speech enhancement method to improve speech recognition in car noisy environments. Noise reduction techniques for speech enhancement have been widely investigated for various application scenarios and noise conditions. Most of these techniques are based on the assumption that noise and speech are uncorrelated. However, in the real world, noise might not always be uncorrelated with speech signal, especially if reverberations exist due to the geometry of the space. The conventional noise reduction methods involving single and dual channels experience insignificant performance improvement due to this unrealistic assumption. Moreover, because of the varying degree of noise being introduced between speaker and microphone, the performance becomes sensitive to the location and directivity of the channel with respect to speaker. The proposed method can be divided into three modules, cross-correlation, channel-selection, and Griffiths-Jim and ANC combination modules, respectively. First, in the cross-correlation module, the input signal becomes output signal with reduced channel noise, particularly burst noise, using cross-correlation function. Second, the primary and reference pair input signal to the Griffiths-Jim and ANC combination module takes in speech dominant signal as primary input and noise dominant input as reference out of the channel-selective parameter. Finally, the aim of the last module is to reduce the background noise using ANC. The enhanced signal produced by the proposed channel selection method gives better performance independent of between channel distance or distance between speaker and microphone.. For example, the conventional dual channel performance evaluation of Griffiths-Jim and ANC hybrid structure shows that the performance is better with the channels being close to each other. In particular, we propose a channel section scheme that establishes primary and reference channel from the initial 2 paired 3-channel inputs. This step essentially establishes the primary channel being speech dominant while the reference channel becomes noise dominant. The established primary and reference pair become the input to the Griffiths-Jim and adaptive noise cancellation (ANC) combination speech enhancement module. The representative experiments show that the proposed scheme is effective in achieving improved recognition in noisy car environment.
※ AI-Helper는 부적절한 답변을 할 수 있습니다.