音声強調を用いた地下鉄アナウンスの可聴性の改善について

摘要

多くの人々が利用する地下鉄において駅名やドアの開閉方向を乗客に伝える車内アナウンスは重要である．しかし，地下鉄走行時に発生する騒音によってこれらの情報は聞き取りにくくなる．音声の帯域を広くとることで音声の可聴性が改善されることが先行研究によって指摘されているが，子音に関してはまだ十分といえる認識率を得ていない．そこで，本研究では地下鉄車内のアナウンスの可聴性を改善する新たな方法の検討を目的とする．まず，可聴性の関係を探るため，騒音と音声のスペクトル分析を行い，音節全体に対して振幅強調を施す必要があることが分かった．次に，4種類のSNRを用いて音声の振幅を調節し，音声と騒音を足し合わせた音声試料を被験者に聴いてもらい，音声試料が何と言っているのかを聞き取ってもらう聴覚実験を行った．聴覚実験の結果，子音の認識率は，先行研究の結果よりも約10％改善された．今後は，母音に対して定常部抑圧処理を施すことで，音質を保った状態で，子音の可聴性を更なる改善を検討していく．%In the public transportation system such as subway, usually it is so noisy that passengers even cannot hear the announcement like what the next stop and which door would be opened clearly. To solve this problem, Obayashi, et. al proposed a method that extends the frequency of the voice. However, it was difficult to get good results for some consonants with the proposed method. Therefore, in this study, we newly propose a method to investigate and improve perceive consonants. First of all, we analyzed some announcements taken inside a subway, and compared the consonants by analyzing power spectra of obtained announcement. As a result, we set up a hypothesis "Emphasizing the whole band of phoneme is the key point of effectiveness". Then, we carried out a hearing experiment in which the person under tested hear the obtained noisy voices and ask them what they heard. From the experiment, we found out that the recognition of consonants comparing with the preceding study has been improved about 10% under keeping the voice quality clearly. As the future subject, we will go over more improvement of each consonant by using suppressing steady-state portions.