|
The Aurix TTS speech technology comes from 20/20 Speech, an equity venture between NXT plc and QinetiQ, which was formerly the Defence Evaluation & Research Agency. The company is releasing the software as a software development kit for programmers to devise their own applications.
The company said its software differs from traditional text-to-speech tools, which create sound by stringing together words and sound fragments drawn from a large database of recorded speech. The traditional method can produce choppy-sounding speech, and for mobile workers the method also demands significant resources, since the speech database is typically stored on a remote server to which the user must constantly be connected over an open phone line.
20/20 Speech says its method is different because it does not attempt to string together fragments of recorded speech to form words. Instead, the company has devised a means of generating sound by simulating the resonant frequencies of the human vocal tract -- essentially, the software can dynamically "create" the sound of a human voice, matching this to a line of text to simulate more realistic speech.
The company said its approach requires far less computer memory and so the intelligence can be stored on the handheld computer or mobile itself, instead of requiring a constant connection to a remote server and its large speech database.
Michael Smith, marketing executive with 20/20 Speech, said that while the end result does not sound exactly like a human voice, it is easier to understand because the speech is not as choppy as the speech from traditional methods. "The speech produced by our method is more intelligible because it's smoother," he said. "The user is more likely to understand what is said."
The technology has been beta tested by companies in industries including automotive and aerospace, the company said, and it is envisioned that the software will be used in applications where users are frequently working in a "hands busy, eyes busy" situation.
Because the speech created by the Aurix software is completely computer-generated, the company can also apply different "voices" to the speech engine, and has devised voices for a UK male and female, a child's voice and a "whispering" voice, Smith said.
The new software follows other speech-based products from 20/20 Speech, including a speech recognition engine and a subtitling tool that was co-developed with the BBC to analyse scripts and apply subtitles to footage automatically.
More information is at http://www.2020speech.com.
|
