• Description of Apple Computer's PlainTalk™ Text to Speech (TTS) Products
• Product Positioning
• Q & A
1. Description of Apple's TTS Products
The Products
Apple Computer has released two products that convert typed text into audible speech, otherwise known as "Text to Speech". These two products are called MacinTalk 2 and MacinTalkPro II. The minimum CPU requirements are: a Macintosh Plus or greater, running System 6.0.6 or later for MacinTalk 2; and Macintosh LC or greater running System 7.0 or later.
Function
Both products perform the same function: They allow a Macintosh to "speak" by turning typed text into audible words. TTS is not another way of accessing stored or recorded sounds and playing them back on cue, rather TTS constructs the speech directly from the typed text by following a set of grammar, syntax, and contextual rules in order to decide how to pronounce the words.
The Two Products Compared
MacinTalk 2 and MacinTalkPro 2 perform the same task, but are very different. There is between a 4:1 to 11:1 difference in the footprint of the two applications, with MacinTalkPro II taking between 700K and 2.6MB, and MacinTalk II at about 150K RAM. Functionally, MacinTalk II is not as sophisticated in its distinctions of the subtleties of the English language. The pronunciation accuracy of MacinTalk 2 will sometimes be less than that of MacinTalkPro 2.
Other things that MacinTalk 2 has more trouble saying in a "human" way than MacinTalkPro 2 are abbreviations, numbers, proper names, and words that don't follow the rules of English grammar or common pronunciation.
MacinTalkPro 2 is much more adept at providing human-like interpretation of words that are unique in their pronunciation or spelling or that do not follow the general pronunciation rules (e.g. "enough", which a simple speech engine may pronounce as "enoog"). MacinTalkPro 2 is also much better at more subtle tasks, such as distinguishing between numbers in currency format as opposed to telephone numbers, and pronouncing them in the way a person would.
2. Product Positioning
What is Apple Releasing?
The two TTS products Apple is releasing are licensed to developers, much like QuickTime. They are intended to allow Apple's developers to create new products or enhance existing products by including spoken feedback from the Macintosh to the user. They will require licensing from Apple if the developer intends to distribute either or both of the TTS products with one of their own packages. The licensing is outlined briefly here. See the licensing agreement for more detail.
MacinTalk 2
This product will come with the TTS manager and can be licensed for $1,000 per developer application per year.
MacinTalkPro 2
This product will be licensed for $1,500 per developer application per year, and will come with MacinTalk 2 and TTS Manager together.
Note: You have to ship the API with your application and the speech engine, no matter which engine you choose to license.
What are Some of the Markets for TTS?
Remote access of data from a phone would allow a user to call his or her Macintosh from any telephone and access data without seeing the screen. DTMF (aka "touch tone") applications with voice feedback allow more flexibility in the design of databases since the person accessing the data gets spoken information without having to find an operator or account representative for help. TTS plays an important role in the ability to access E-Mail remotely, remote database access, and other DTMF applications (telephone-controlled data access/manipulation similar tele-banking). MacinTalk 2 and MacinTalkPro 2 will bring the benefits of TTS currently enjoyed only by large companies like telephone companies, banks, and insurance companies to individuals and small businesses.
TTS is essential for products designed for disabled or blind computer users who have trouble reading the screen or speaking. There is already sizable market demand as a result of these needs.
In existing applications, this could take the form of allowing the user to receive system prompts, messages and cues in English, rather than in written form or in beeps and boings.
Questions & Answers About Text To Speech
Who is the product for?
MacinTalkPro II and MacinTalk II are developer products, like QuickTime. They are not stand-alone retail software products.
Is there a retail version of either product?
No, the product is currently available only as a developer product.
Which product should I license if I am interested in developing a product that includes TTS?
That depends on what is more important to you: Memory space or performance. The cost of licensing the products is not significant. If you want a small application (MacinTalk 2 is about 150K) and are willing to give up some of the performance that MacinTalkPro II (between 700K and 2.6MB) offers, then MacinTalk 2 may be the way to go. However, since MacinTalkPro 2's TTS is so much more "human" in its interpretation of text, it may be worth the extra RAM space. Again, the right choice depends on your market/product needs and constraints.
How will the products be distributed?
Both MacinTalkPro II and MacinTalk II will be initially distributed in three ways.
1. Sent to all developers free on the Developer CD ROM.
2. Available for licensing to developers
3. Available for sale by developers.
What will it cost me to license the products?
It will cost $1,000 for MacinTalk 2 or $1,500 for both MacinTalk 2 and MacinTalkPro 2 (per application) per year.
How does Text To Speech work?
In layman's terms, the software is embedded with the rules of pronunciation of the English language. When it is instructed to "speak," it first scans the text and converts any non-text items (i.e. numbers, dates, abbreviations, etc.) into text. For example, it would change Dr. to "Drive" or "Doctor" as appropriate, 1987 to "nineteen eighty seven", and $19.87 to "Nineteen dollars and eighty-seven cents." This step is called normalizing.
Next, the computer looks for any words that are anomalies (like "of" which is the only word that ends in "f" where the "f" is pronounced "v"). After it has looked up all of those and determined the correct pronunciation, it then looks for the words that have multiple pronunciations depending on context ("record" or "read" are common examples).
The software then performs several steps to convert the letters to "phonemes" which are essentially sub-syllables. The final steps are to add a cadence to the speech so the pronunciation is not a pure monotone, and to fine-tune the pronunciation and emphasis of certain syllables or phonemes that exhibit subtle differences depending on their context.
Once all of that has been accomplished, it can begin to convert the text into spoken English by sending the information to a speech synthesizer.
How long will it take me to create a product that incorporates TTS?
Because our products are designed to work with a standard API, it is very easy to integrate TTS with a standard Macintosh application. Most developers who took part in our beta seeding program were able to incorporate TTS into their applications in less than a week.
Can either of the products speak in a foreign language?
Not now, though Apple is working on this important need. Each localization requires a great deal of work since different languages can have entirely different rules for syntax, grammar, alphabets, and pronunciation.
Didn't Apple have a Text To Speech product on the market a long time ago?
Yes, it was called "MacinTalk," but it was very primitive and required phonetic spelling in order to speak with anything resembling human pronunciation. Furthermore, it did not do the steps outlined above to normalize the text or make pronunciation decisions based on context or location in a sentence. Although the original MacinTalk is still shipped with some third-party Macintosh products, Apple has not supported it for years. Apple will support both MacinTalkPro 2 and MacinTalk 2 for the foreseeable future.
