Oleg,
There are my answers below.

>although the usage of SPI might be an idea
Only the SPI in the SLAVE mode that's important for this application.

>Some Atmel derivatives, for example AT89C51RD2,
>require for that the maximum SCK frequency allowed
>in slave mode must not be above 1/4 of peripheral clock.
OK, but where is a problem with SPI clock? Original poster needs only 19200 baud to get words from vocoder! Btw, I referred to the AT89LS53 chip, why we discuss about AT89C51RD2?

GM>2) Use the channel receive data strobe for synchronization
GM>of receiving the new byte (the first half of the 16-bit word).

>Maybe. But this signal is active high and what is the bad as well:
>it may ends at any level of CHAN_TX_CLK so SPI mode is hardly defined.
Definitely I don't understand what are you talking about. I said: just take this channel transmit data strobe from the vocoder, connect it to the external interrupt pin and use 1-0 (or 0-1) transition for synchronization of the first byte of the frame of a vocoder.
The quotation from AMBE-2000 user's manual says:

1) Wait for slightly less than 20 msec.
2) Assert CHAN_TX_STB and read word on CHAN_TX_DATA.
3) If transmitted word not 0x13EC, discard it and repeat step 2.
4) If transmitted word 0x13EC, read 23 more words (rest of packet).

Perhaps it is clearer now.

>Well, datasheet of AT89S53 does not indicate next
>"feature" of SPI called Overrun Condition...
I was not talking about the Overrun Condition of the MC but only about the OVERHEAD parts of received packet witch consists from 12 words of overhead and 12 words of data.

>What if some other interrupts at the same or high priority level "eat" this time.
You know what to do: put this interrupt into the highest level, that's all.

>Maybe AT89S8253 may helps then with its double-buffered SPI.
Could be. My suggestion is only an idea, only one possible way how to implement hardware SPI for the vocoder specified above.

Regards,
George