11 Speech

The TI-99/4A had, back in the day, perhaps the greatest speech capabilities of any of the plethora of "home computers" that were so prevalent during the early 1980s. Coupled with the Solid State Speech Synthesizer it was pretty much unbeatable.

If you have a speech synth connected, you can take advantage of it in TurboForth. Unlike languages such as Extended Basic (which also offered speech synthesizer access) your program can continue to execute while the speech synthesizer is talking.

11.1 Types of Speech

The Speech Synthesizer supports two types of speech:

Speech from a built in vocabulary of words using a male voice;
Streaming of raw (Linear Predictive Coded) data directly into the speech synthesizer, allowing any word to be spoken in any voice (provided you have the data)

TurboForth supports both methods.

11.2 Speaking Words from the Speech Synthesizer Vocabulary

The word SAY is used in TurboForth to speak words from the speech synth's built-in vocabulary. The stack signature for SAY is as follows:

    SAY ( address count -- )

As can be inferred from the above stack signature, SAY actually expects an address of a list of data (address), and the number of items in that list (count). The word DATA can be used in TurboForth to build lists.

Within the data list itself consists of the number of data items, and the data items themselves. So what are these data items? Well, in the case of SAY, they are simply the addresses of the words in the built-in vocabulary.

Example

Let's try an example. Let's make the speech synthesizer say "I LIKE FORTH":

    : ILF DATA 3 $3793 $3F2F $2D19  SAY ;

Here, we have set up a data list consisting of three values. These values are the addresses of the words in the synthesizer vocabulary:

"I" - address 0x3793
"LIKE" - address 0x3F2F
"FORTH" - address 0x2D19

At run-time (when execute the word ILF) DATA pushes the address of the start of its data-list, and the number of items in the list (the count) to the stack, which is exactly what SAY expects.

If you run the above example (just type it in, then type ILF) you'll hear the speech synthesizer tell you in its rather gruff deep male voice that it likes Forth!

*Note: If you are running TurboForth in a PC based emulator, then the best emulators to use for speech are MESS and Win994A. Classic99 is not currently recommended for speech.

11.3 Speaking Using LPC Data

If you have the LPC data, or the means to create it, you can make the synthesizer say anything you like. The technique is exactly the same: Use DATA to define the list of bytes to send, and then use the word STREAM to stream the data into the speech synthesizer in the background.

Example

Here is an example which says "Forth is Awesome!"

HEX : FIAd DATA 86 6004 08D2 8D00 D986 09A0 FA0C 0364 1766 0056 DC4B C5F2 2937 276A B518 2B1A DC7A 35A6 2E65 72AB D3F8 B81C 4DAB 4E9B 9A49 05B5 3C6D 5EA6 25BC F034 6514 A771 9DD3 643E E411 B514 4079 B800 3253 5640 9626 ABCE 96D9 CBDB 9C3A 6991 E8EA 74EA AC45 6D73 D569 B212 8BA9 CEA7 6E46 D822 BA9E B6B1 1071 EFBC 9A49 5D95 BC49 002A A048 8093 1BA7 89A1 D8AC 139F 36E5 6197 4C7C 9A14 4E4D 33C9 69B2 7E37 F126 A7C9 EECD B992 9C26 DB37 974C 739A 54CE D45B F569 1A7B 0B2E BBAB 5DDA D9C2 6C07 6031 E200 5CC3 EC80 632A 1D90 81E5 99B3 2B74 8E38 67A8 AAC9 859B ACB1 9821 339A DCA6 E231 9DC2 499B 4A24 4B32 2763 AA89 2CC1 9D8C B926 0AC7 9093 E612 300D 53A9 59B2 456F CDB8 6AD1 4AB2 2C63 C303 0F0F ; DECIMAL \ stream speech data to the synth & give it some work to do ) : FIA ( -- ) FIAd STREAM 128 33 DO 0 0 I 960 HCHAR LOOP PAGE ." Forth is awesome!" CR ;

It is taken from Block 2 of the TurboForth companion disk, which contains a number of demo's and utilities, all in source code form.

When using DATA with long streams of bytes (anything that is more than 80 characters in length) then the source code should be entered via a disk block. If you have the companion disk, you can try the above demo by typing 2 LOAD

11.4 Background Speech

Interfacing with the speech synthesizer is non-trivial. Care must be taken to give the speech synthesizer time to executes its commands, and the input FIFO must not be allowed to overflow, otherwise garbage/garbled speech is the result. The words SAY and STREAM use interrupts to 'drip feed' the speech synthesizer with data. However, it should be noted that interrupts are only breifly enabled and then disabled again in TurboForth during the operation of words that accesses video memory. Consequently, for speech to function properly it some type of video access should occur regularly. The reason it is possible to execute speech words from the command line is that the cursor is flashing, which causes the VDP interrupts to fire, which gives the speech driver an opportunity to execute and service the speech synth. In practice, this is isn't a big problem, as speech tends to be used in games, where reads/writes to the video memory are of quite a high frequency, thus giving the speech driver plenty of opportunities to execute. The alternative would be to enable interrupts all the time, however this would have the un-desirable side-effect of slowing down the main execution thread (your Forth program).

11.5 Determining Speech Synthesizer Status

It is possible to poll the speech synthesizer and see if it is busy (talking). Simply use the word TALKING?

    TALKING? ( -- true/false)

As can be seen, TALKING? returns either true or false. True means the speech synthesizer is busy (talking). This word should be used when you need to send successive speech phrases to the synthesizer; as you will want to ensure that one phrase has finished before the next one starts:

Example

The following program says "I like Green", then it says "I like Red". In between the two phrases, it waits to allow the first phrase to finish before starting the next one.

: GREEN DATA 6 $3793 $3F2F $321D $3793 $3F2F $321D SAY ; : RED DATA 3 $3793 $3F2F $57C1 SAY ; : WAIT BEGIN 0 V@ DROP TALKING? NOT UNTIL ; : TEST GREEN WAIT RED ;

Note that WAIT, as discussed above in 11.4, access video memory (it does a read from VDP memory address 0 and then simply drops the result, as it is not needed. This gives the VDP interrupts a chance to executes, which in turn allows the speech driver to run, which feeds the data to the synthesizer.

The loop in WAIT simply checks the synthesizer by calling TALKING? and loops until False is returned by TALKING?. The word NOT in the loop simply inverts the result returned by TALKING?, since UNTIL loops until the result passed to it is true. Since we want the loop to exit on a false, we invert the result returned by TALKING?.

An alternative, (though slightly slower) version would be:

    : WAIT   BEGIN  0 V@ DROP  TALKING?  FALSE = UNTIL ;

Where, as can be seen, the loop repeats 'until false'. I.e. Until TALKING returns false. Here, the result from TALKING? is compared to FALSE using the word = (i.e. is the result returned from TALKING? equal to FALSE).

11.6 Determining if Speech is Supported

It is not necessary for your Forth program to determine if the speech synthesizer is fitted to the machine. In the event that the speech sythesizer is not fitted, the speech driver aborts and returns immediately. Therefore your Forth code can simply assume that speech is available and the system will take care of the rest.

<-- Back to Tutorials