| 摘要 |
1,246,079. Speech analyser. INTERNATIONAL BUSINESS MACHINES CORP. 27 May, 1970 [2 June, 1969], No. 25406/70. Heading H4R. In a speech analyser rectified signals are derived representing the envelope of the speech components in respective frequency bands distributed across the speech spectrum. The rectified signals are summed and applied to a frequency discriminator to derive a voltage level proportional to the pitch frequency of speech during voiced sounds and a noise signal during unvoiced sounds. The level of high frequency components in the signal is used as a voicedunvoiced signal. As shown, Fig. 2, speech from microphone 1 is amplified and fed to a bank of filters 3 each having a bandwidth of about 300 Hz so as to pass two pitch harmonics. The filter outputs are rectified in 4 combined in a summing network 5 passed through a 70-250 Hz band pass filter 6 and modulated to about 15 KHz in balanced modulator 7. The upper sideband of the modulator output is selected in filter 8 and the signal limited and filtered to provide, at the output of filter 10, a constant amplitude signal of average frequency related to the pitch frequency of the speech, which is passed to frequency discriminator 11. The output of the discriminator is a signal whose level is proportional to the pitch frequency during voiced sounds and is noise during unvoiced sounds. The relatively slowly varying pitch signal is fed through LP filter 12 to output V1 while the noise signal is fed through HP filter 17 and rectified to provide a signal of one level, sufficient to operate threshold circuit 20 during unvoiced sounds, and a lower level signal, insufficient to operate 20, during voiced sounds. |
