| 摘要: | 本研究之目的在於探討以麥克風陣列和噪音消除策略為前處理的人工電子耳雙耳聽模擬在噪音環境下對中文語音辨識率之影響。本論文中人工電子耳模擬是採用進階結合編碼策略(Advanced Combination Encoding, ACE)，助聽器則是以一個截止頻率為500Hz的低通濾波器模擬，前處理的部分包括兩種不同的麥克風策略(Mic.)和兩種不同的麥克風策略加噪音消除策略(Mic.+NR)，對十位聽力正常成年人(九男ㄧ女)做語音辨識率的測試。實驗中所使用的訊噪比分別為 -5dB和0dB，語料則有雙字詞與句子兩種加語音噪音。受測者進行單獨人工電子耳和人工電子耳結合助聽器的模擬實驗。在單獨人工電子耳的情況下，語料有經過前處理(雙字詞:50%，句子:42%)和沒有經過前處理(雙字詞:17%，句子:22%)的語音辨識率之間有顯著差異(p <0.001)，同樣的在人工電子耳結合助聽器的情況下，語料是否經過前處理(雙字詞:93%、61%，句子:97%、79%)的語音辨識率之間也存在著顯著差異(p <0.001)。而人工電子耳結合助聽器和單獨人工電子耳在有經過前處理的雙字詞和句子實驗中平均辨識率呈現出顯著差異(p <0.001)。另外在麥克風策略和麥克風策略加噪音消除策略之間的辨識率，不論是在單獨人工電子耳(Mic:47% , Mic.+NR:44%)或是人工電子耳結合助聽器(Mic:95% , Mic.+NR:95%)的實驗中都沒有顯著差異(p > 0.05)。由上述可以得知，不論是在單獨人工電子耳或是人工電子耳結合助聽器的情況下，有經過前處理和沒經過前處理的語音辨識率之間存在著顯著差異，但是增加噪音消除策略主要是增加聆聽舒適度而非改善語音辨識率。 The purpose of this study is to investigate the pre-processing with microphone array and noise reduction for electroacoustic stimulation of cochlear implant (CI) simulation on Chinese speech recognition in noise. We used advanced combination encoding as the speech processing strategy to simulate the cochlear implant and a low-pass filter with the cut-off frequency of 500 Hz for the hearing aid (HA). Four pre-processors (two different microphone arrays, Mic. and these two microphone arrays plus two noise reduction strategies, Mic.+NR) were implemented in this study. There were 10 adults (9 males and one female) with normal hearing participating in the experiment. We used disyllabic words and sentences combined with speech-shaped noise (SSN) as test materials (signal-to-noise ratio, SNR of -5dB and 0dB) in all experiments. All the subjects attended two experiments: one is CI only and the other, CI combined with hearing aid (CI+HA). In the CI only experiment, average speech recognition rates of the test materials (word and sentence) with pre-processing (word: 50%, sentence: 42%) and those without pre-processing (word: 17%, sentence: 22%) were significantly different (p<0.001). Additionally, there were significant difference (p < 0.001) between the average speech recognition rates of the test materials with pre-processing (word: 93%, sentences: 97%) and those without pre-processing (word: 61%, sentences: 79%) in the CI+HA experiment. The average speech recognition rates between CI+HA and CI only showed significant differences (p <0.001) between CI+HA and CI only irrespective of the test materials and the pre-processing strategies. However, average speech recognition rates between pre-processors with microphone arrays and microphone arrays plus noise reduction strategies showed no significant differences (p > 0.05), no matter under the experiment of CI only (Mic.: 47%, Mic.+NR: 44%) or CI combined with HA (Mic.: 95%, Mic.+NR: 95%). In summary, our results implied that there are significant differences between the speech recognition rates of the test materials with pre-processing and those without pre-processing in both experiments. However, our data also showed that noise reduction strategies are used to improve listening comfort for the subjects instead of improving speech recognition rates. |
