A spoken digit recognizer for Japanese has been designed and constructed. A set of eight features (concerned chiefly with the formant structure) which appear to characterize the individual spoken digits most efficiently has been determined by examining sound-spectrograms of typical utterances. Recognition is performed based on the Bayes decision rule applied to this set of features. The extraction of the features and the decision-making are carried out by fully transistorized circuits. Although the circuits are relatively simple, results of 99.7 percent recognition for 1000 utterances of one male speaker and 97.9 percent for 1000 utterances of 20 male speakers were obtained.
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.