摘要 |
1,055,371. Automatic character reading. STANDARD TELEPHONES & CABLES Ltd. Jan. 8, 1965, No. 963/65. Addition to 1,012,765. Heading G4R. In speech recognition equipment information is extracted at 5 (Fig. not shown) from each of a number of speech inputs 1 sampled in turn by gates 3 and the information is stored at 6 reference data stored at 7, compared in unit 5, a recognition signal causing a corresponding standard word recorded on magnetic drum 11 to be set back to the corresponding input. Each input is connected to a squaring circuit 2 and the output of each is sampled at high speed by the multiplexing gates 3. The circuit 5 may derive information from the incoming signals as described in the parent Specification by measuring the intervals between zerocrossings by counting reference pulses. The data from each input is stored at a particular location in store 6 and as results of further samplings are received this stored data is withdrawn, up-dated, compared with the reference information in store 7. If there is insufficient agreement for recognition the data is returned to store 6. If there is recognition the recognition outputs on leads 8 and 9 identify the word and the input channel on which it was received. These signals are applied to control unit 10 to cause the corresponding word to be read from the magnetic drum 11 and sent by switching unit 12 to the appropriate input line. The speaker thereby hears the word recognized. This serves both as a check on the recognition and as an illustration to the speaker of the proper pronunciation of the word. |