Method and device for eliminating the disturbing influences of the fundamental frequency of the voice and its harmonics in the frequency channels of a speech coder

摘要

1,120,706. Vocoder. INTERNATIONAL BUSINESS MACHINES CORP. 15 April, 1966 [26 April, 1965], No. 16552/66. Heading H4R. A channel vocoder which includes a number of band-pass filters to derive signals defining the spectrum of a speech signal has an integrator arranged on the output of each band-pass filter with the integration time of each integrator controlled in accordance with the pitch frequency of the input speech signal. The arrangement eliminates pitch frequency dependence of the various analysis channel outputs. Fig. 2 shows a system in which the speech spectrum is analysed by band-pass filters BP 1 to BP n whose outputs are rectified and integrated in circuits I 1 to I n to provide signals defining the speech spectrum. The integrators are sampled and reset by pulses on line K o , the pulses being derived from the speech pitch pulses during voiced sounds and from a constant repetition rate pulse generator IG during unvoiced sounds, switching being under the control of a voiced-unvoiced discriminator D1. In a vocoder where the excitation function is derived as a pulse train, both during voiced and unvoiced sounds, the pulses being randomly spaced within limits during unvoiced sounds, the excitation function pulses may be used to sample and discharge the integrator both during voiced and unvoiced sounds, Fig. 3 (not shown). Where the spectrum signals are to be transmitted in digital form an arrangement such as is shown in Fig. 5 may be used where the integrator has a threshold discriminator D 1 to D 1 connected to reset the integrator when the integrator output reaches a predetermined value. Each resetting produces a pulse on line e and these pulses are counted by the respective counter BZ 1 to BZ n . Each excitation function pulse is applied to store the counter count and to reset the counter to zero. If the spectrum signals are to be quantized, e.g., to eight levels, the integrator and level discriminator may be arranged so that it produces a maximum of 8 pulses on line e for the maximum signal level during the longest expected excitation function period. By using a three-stage binary counter for each counter BZ 1 to BZ n a binary signal is obtained directly from the state of the binary counters. Excitation function pulses may be derived as shown from the speech signal during both voiced and unvoiced sounds, or the resetting pulses may be derived from an idependent generator during unvoiced sounds as in Fig. 2 (or Fig. 4, not shown). Transistor integrator circuits are described, Figs. 6 and 7 (not shown).