This paper describes a recently developed adaptive speech-recognition system. The system quantizes the spectrum of the speech signal in respect to frequency, amplitude, and time. A 20 bit binary feature matrix is obtained for each utterance. Each feature matrix, identified as to the word spoken, may be stored in a disc memory. With the system in its present form, up to 256 reference samples for 10 classes of words may be accumulated. When the desired number of reference samples have been obtained, the system may be used for the recognition of speech. Test words are spoken and quantized as before. The new feature matrix is compared serially with the contents of the disc memory. The degree of difference between each reference sample and the test sample is determined. If this difference is sufficiently small, the identifying information on the reference sample is decoded and a visual display actuated. This system, with the aid of the operator, adapts itself in an optimum manner to the characteristics of the speaker -training- the system. The contents of the memory may be readily changed for a different speaker or vocabulary.
https://www.aes.org/e-lib/browse.cfm?elib=923
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.
Learn more about the AES E-Library
Start a discussion about this paper!
