German Text to Speech
What do you think of when you hear the term "synthetic" or "computer" speech? If you thought immediately of an expressionless, robotic, monotone voice, you join the majority of people who have heard computers "talk." But what if you could get your computer to read any German text it comes across-in a natural-sounding voice (male or female) with accurate German pronunciation? Wouldn't that be a cool way to help you learn German?
That may not have been the main inspiration for the developers at AT&T Labs, but the result of their efforts is a synthetic-speech engine with artificial voices for several languages that sound very human and far exceed anything I have ever heard from a computer before now. (HAL in the film 2001 doesn't count; HAL's computer voice was portrayed by actor Douglas Rain.)
It may surprise you to learn that AT&T (Bell Labs) first worked on speech synthesis as far back as the late 1930s. Even earlier, around 1870, the German scientist Hermann Ludwig Ferdinand von Helmholtz (1821-1894) and others studied the relationship between the sound spectrum and the human voice. But the technology for creating electroacoustic speech did not arrive until the invention of the telephone. Now in the age of computers, AT&T has dubbed its latest text-to-speech (TTS) technology Natural Voices™ and offers it to software developers to create computer programs that use AT&T's synthetic voices. So far, there are Natural Voices™ for English (American and British), French (Parisian), German, Spanish (Latin American), and even Latin. Each language set includes both male and female voices. (You can take a "test drive" with the AT&T Natural Voices™ Demo on the Web.)
If you think about it, getting a computer to read text is not as easy as it may seem. There are many words and abbreviations in any language that can be pronounced in more than one way, depending on the context ("bass," "live," "read" or "Dr." in English). AT&T has built in enough sophistication for the desktop version of its TTS engine so that if it encounters a sentence with abbreviations, as in "Dr. Smith lives on Ocean Dr.", it can determine that the first "Dr." should be pronounced "doctor" and the second "Dr." should be pronounced "drive." Or take the English word "read," which can be pronounced either as RED or REED. The Natural Voices engine can correctly pronounce "has read" versus "to read." Natural intonation (in a question, for instance) is another problem, as are stress, timing and pitch. Although it isn't perfect, the Natural Voices engine usually performs well in these areas, in both German and English.
For German, Natural Voices uses two 16K synthetic voices named Reiner and Klara. For the desktop version, they must be purchased (on CD or as a download) from AT&T or as part of licensed software that uses the AT&T technology. I have tried one text-to-speech program (PleaseRead™) that uses both the Natural Voices engine (with Reiner and Klara, purchased separately) and German voices from Microsoft. From what I heard, the Natural Voices German synthesized speech is far superior to the German Microsoft voices. If you want to know more about the PleaseRead software, read my upcoming review.
The pronunication performance of Natural Voices can be tweaked by the user, using custom dictionaries to modify the sound for particular words or phrases. For instance, although the Natural Voices TTS engine already knows how to pronounce the names of any of the Fortune 500 firms, you can add other brand or company names to the language-specific pronunciation dictionaries.
The Natural Voices desktop software is only available for Windows. It works best with a newer, speedy computer with Windows XP and lots of memory, but in the ReadPlease program it worked fairly well (with an occasional freeze-up) when I tested it on an older 800mHz computer running Windows 98 with 256MB of memory. But anyone can try Natural Voices online at the AT&T demo Web site.
MORE > Software for German
Natural Voices is a registered trademark of AT&T Corp.
Windows is a registered trademark of Microsoft Corp.
ReadPlease is a trademark of ReadPlease Corporation
Apple's Mac OS X Text-to-Speech
|Apple's Mac OS X platform has a built-in text-to-speech engine that can read a selected English text. To use this feature, you have to select it in System Preferences under "Speech." If you have your language preferences set to "Deutsch," it will read German. It's not quite as good as AT&T's Natural Voices, but it's free as part of OS X.|