General Constraints
D-2
D.1 General Constraints
The sound quality of the speech produced by the MSP53C391 and
MSP53C392 is sensitive to the timing of the waveforms which transfer speech
data to the device from the master microprocessor. Depending upon the algo-
rithm being used to synthesize the speech, the speech data is stored in a circu-
lar buffer either 16 or 32 bytes wide. If the data is written to the slave too infre-
quently, the buffer will empty and the synthesis process will stop or be cor-
rupted. If the MSP53C391 or MSP53C392 is polled too frequently to determine
whether or not it is ready to accept new data, then too many of the internal
instruction cycles may be used servicing the polling process and the sound
quality can be degraded.
Each of the synthesis algorithms tends to use the data in the buffer in
bursts
followed by a period of time in which the data is being utilized before more data
is read from the buffer. Once the MSP53C391 or MSP53C392 has been polled
and determined to be ready to accept new data, the data should be loaded
quickly until the buffer is again full. The read pulses should subsequently be
spaced more widely until it is again determined that the buffer is not full and
the device is ready to accept new data.
The spacing of the read pulses while the buffer is an important determinant of
the synthesized speech quality, but is difficult to specify precisely due to the
different data throughput requirements of the different algorithms. In general,
the optimal polling frequency will increase with the bit rate of the synthesis al-
gorithm being used. In the sections below are some timing waveforms which
we have found to work with the datasets that we have tested. In many cases
it will be difficult for the system designer to exactly replicate the timing shown
in these sections. The timing should be adjusted to optimize the sound quality
for the specific system being designed.
The following general considerations should be observed:
Keep the STROBE pulses as short as possible, but they should not be
shorter than that shown in the waveforms below.
Once it is determined that the buffer is not full, load new data quickly.
Once it is determined that the buffer is full, read the status of the BUSY
signal periodically, but not as frequently as when refreshing the buffer with
new data.
Содержание MSP53C391
Страница 1: ...MSP53C391 and MSP53C392 Speech Synthesizers User s Guide May 2000 SPSU016A Printed on Recycled Paper ...
Страница 4: ...iv ...
Страница 62: ...C 1 Appendix A Listing of FMequM2 inc Topic Page C 1 Listing of FMequM2 inc C 2 Appendix C ...
Страница 78: ...E 1 Appendix A Listing of FM2INTR1 inc Topic Page E 1 Listing of FM2INTR1 inc E 2 Appendix E ...
Страница 98: ...F 1 Appendix A MSP53C391 and MSP53C392 Data Sheet Topic Page F 1 MSP53C391 MSP53C392 F 2 Appendix F ...
Страница 99: ...MSP53C31 and MSP53C32 Data Sheet F 2 F 1 MSP53C31 and MSP53C32 Data Sheet ...