摘要 |
PURPOSE:To make a pseudo voice signal resembling to a voice signal in an actual environment. CONSTITUTION:A Hoth spectrum noise indicating a room noise 37 of a speaking environment 31 is added to a high quality voice signal, typical overall frequency characteristics of a microphone 34 and a telephone channel 35 and band limitation of 300Hz-3.4kHz are given to the signal in which a noise is added, after that, a white noise imitating a channel noise 39 is given and a pseudo voice signal is obtained. By using this signal, a voice model is learned with a HMM method, and an unknown telephone voice is recognized using the model. |