| 摘要 |
1470438 Vocoders WESTINGHOUSE ELECTRIC CORP 22 April 1974 [8 May 1973] 17510/74 Heading H4R A formant tracket includes tunable electronic filters 20, 30 having respective frequency responses (such as A, B, Fig. 5) and whose outputs are compared 40-44 to produce a control signal which adjusts the frequency responses until the outputs become equal at the formant frequency F, Fig. 6. The arrangement of Fig. 1 forms part of a cascade to extract the pitch frequency and the first three formants. The filters initially centre on f 0 - #f and f 0 + #f respectively and therefore produce different outputs when a formant F, Fig. 5 occurs. The filter outputs are rectified 40, 44 in opposite senses, added 42 and integrated 50 to control FET's 23, 33 to shift the centre frequencies to the Fig. 6 position. The Q-values of the filters may simultaneously be raised by controlled FET's 24, 34 to reduce the bandwidth. The lowpass output from filter 20 is added to the control signal to bias the circuit to detect the lowest frequency formant in the band. The integrator output is gated 60 to processing circuitry S when the band output output from filter 30 exceeds a threshold to exclude the beginning and end of a response from transmission. Amplifiers 67, 68 and differential amplifier 70 prevent passage of the analysed band to succeeding stages. The processing circuitry S may normalize the formants and compare them in a semiconductor matrix to identify a speaker and/or word in a sonogram. |
