| 摘要 |
1,261,385. Speech recognition. MATSUSHITA ELECTRIC INDUSTRIAL CO. Ltd. 9 July, 1969 [24 July, 1968; 27 May, 1969], No. 34692/69. Heading G4R. Speech analysing apparatus generates a signal having a frequency depending on the difference in frequency between an input voice and a standard voice signal, and uses the generated signal to normalize the frequency of the input voice, the normalized input voice signal being split into frequency bands to give amplitude signals which are compared to locate formants the locations of which are stored in their order of occurrence. The speaker speaks the five vowels in turn, the speech waveform in each case being fed via a low-pass filter and integrator to a Schmitt trigger circuit, pulses from which are gated to a counter to determine the pitch, the count being converted to an analogue voltage which is subtracted (in a differential amplifier) from a standard voltage for the vowel, the result being converted to digital form and stored in a memory. Logic circuitry obtains the average of the five results thus stored which is converted to analogue form and used to control the frequency of an oscillator the output of which is used to shift the frequency of the input speech waveform to be recognized, after the latter has been low-pass filtered. This shifting compensates for different speakers. The resulting normalized speech is split into frequency bands. Adjacent bands are compared (in differential amplifiers), after rectification and integration, the amplifier outputs feeding NAND logic via thresholders detecting positive and negative levels respectively, to locate the formants. The formant locations are entered into columns of a core matrix addressed in turn by a driver circuit started by detection of the onset of speech. Different columns may be addressed for different lengths of time. The matrix may also receive voiced and unvoiced indications obtained by integrating high and low frequency bands of the normalized speech separately, comparing the (two) integrator outputs in a differential amplifier and thresholding the amplifier output to detect positive ("voiced") and negative ("unvoiced") outputs. |
