Hi Jon. There are literally thousands of devices out there that can help
your friend. Most of them over priced in my opinion. It very much
depends on several things. Firstly are your friends problems simply
related to speech, if so you can ignore the devices that have
sophisticated input devices for people with physical problems. Secondly
do you want a system that can say anything, or something with just a few
helpful phrases. Or both, usually different technologies are used for
each. Thirdly does your friend have any speech or vocalisation capability?
The sort of devices available include:-
Semantics processors that use graphical displays and icons who meaning
is both context and sequence sensitive, ie an icon of the sun may mean
outside/hot/burning etc depending upon what icons preceeded it and what
follow. These are usually expensive machines, design primarily for
people with slow input capability, making their speech rate faster.
Next come phrase based systems, these typically have banks of phrases
stored that allow sentences to be constructed, often these will have
some graphical interface, or overlay cards, so that you can select from
'sets' of words and phrases, making it unnecessary to have a key for
every one.
Finally there are the text entry type systems, some of these are
literally just type and talk affairs, with a keyboard and a speech
synthesiser. others are a little smarter and include features like type
ahead, where the most frequently used words are prestored. The downside
to this type of system can be speech quality, although the new Winbond
WS071 processors are streets ahead of the old robotic voice synhts of
the past.
I've built most types at different times, but most of my designs were
aimed at people with physical issues as well as speech problems, so were
typically restricted vocab designs. With many of these you can use an
analog voice recording chip, such as th ISD series from Winbond, and get
up to 16 minutes (last time I looked) of reasonable quality speech. By
programming this with common words and phrases you can get quite
apowerful system. On average (based on lots of TV recording and word
counting) a word, without silence, consumes 0.4 seconds, if you assume
that you can get 120 words stored per minute a 16 minute device can
offer quite a vocabulary, call it 1800 words, or about half the number
of words used frequently in conversation, and about a third the
vocabulary of most people, hence, with time, and judicious word
selection you could have human voice speech with a reasonable
vocabulary. You would then have a text entry system using a micro, and a
filing system to construct sentences with.
So two choices as I see it.
First option the Winbond ISD5216 speech storage system, add a micro with
enough grunt to work a file system and control the ISD part, add a
keyboard and a small display to see what is being entered, use 'type
ahead' or 'smart typing' to speed up data entry. This gives human
recorded voice, but is a lot of work entering the vocabulary. You can
enter it incrementally, but that's still a lot of work. Upside, you can
try and match voice and accent.
Second option use a text to speech processor like the Windbond WS071,
available in male and female voices. Then add a micro/keyboard/display
as above, perhaps type ahead, very similar hardware choices whichever
option you choose. This option gives an unlimited vocabulary, it has an
onboard 'dictionary' where you can teach it to correctly pronounce words
that synths find difficult, and quality has become quite good. Downside
is that it still isn't quite human voice.
There is a third option, which is to buy a commercial unit, but these
may be more limited than the solutions I describe, or extremely expensive.
Whichever option, if indeed you choose one of the two hardware
solutions, there is a lot of work involved. The keyboard, display and
type ahead stuff is generic, the synth requires an SPI interface, and
implementation of the reasonably simple protocol. The speech recording
system requires a lot of file system mainrenance software, and a lot of
sentence construction software to be written.
Both of the systems I describe are fully portable. You could implement
the text to speech system on a PC using the parallel port, but this
would tie you down.
The biggest expense in doing it yourself is the PCB. Rather than a
complex 32 key keypad I would opt for a simple 5 key data input. These
can be extremely fast with practice. I use one when I'm bedridden, that
I built myself, fits in the palm of my right hand, one thumb button, and
one under each finger tip, the thumb steps between character sets, mine
is for PC data entry so includes punctuation, this one wouldn't, just
the alphabet, space, numbers, backspace talk now etc. A nice clear large
character 16 x 2 LCD, cheap and adequate, an LM386 for the speaker,
digital volume control, and that's about it.
Al
