| 摘要 |
1316204 Speech synthesizer INTERNATIONAL BUSINESS MACHINES CORP 2 March 1971 5739/71 Heading H4R Frequency spectrum generator, particularly for synthesizing speech, comprises a data store for storing digital samples of a periodic waveform taken at sampling intervals smaller than the waveform period, and a control device operable to read-out the samples at the original sampling intervals or multiples thereof. As described the data store 2, Fig. 3, stores in respective rows the digital values of samples of respective different frequency waveforms, each sample represented as the change in value with respect to the previous sample. The rows feed respective accumulators 8-15 to provide the output signal of the selected frequency according to the address stored in 5. The rate at which the signal samples are fed from store 2 is governed by the division table 1 under the control of address store 4. The division table 1 receives clock pulses from generator 19 and can then supply pulses at the original sampling rate of the samples in store 2, or at any one of a range of predetermined submultiples thereof, to address increment 7 to cause the reading out of samples from store 2 at the appropriate rate. In addition the amplitude of the signals fed out is controlled according to the information set into an address store 6 which, via an amplitude store 3 controls the accumulators 8 to 15 to provide the appropriate amplitude weighting. Phoneme counter 20 is provided to cause new information, in respect of the frequencies to be generated, to be fed into the address circuits 4, 5 and 6, after every 25 Msecs. Pitch counter 21 is used to provide smooth transitions between adjacent speech sounds by superimposing some components of one speech sound on the following speech sound by controlling the address circuits to read out both signals for a short time on transitions. A total of 5 frequencies may be generated simultaneously at any one time, three relating to a new speech sound and two relating to a previous speech sound to be superimposed on a new sound during the transition period between the two sounds. An alternative system, Fig. 4, uses a functional memory 32 which may be of the type of Specification 1,186,703 to perform the functions of division table (1 of Fig. 3), address incrementer (7), accumulators (8 to 15), combiner (16), amplitude store (3) and phoneme and pitch counters (20 and 21). The organization of the functional memory is described with respect to Fig. 5 (not shown). |
