Text to speech is getting emotional
In the early days of text to speech (TTS), the requirement was just that the listener could understand. One of the best known examples is Professor Stephen Hawking, the author of A Brief History of Time, who has used a speech synthesiser for many years that sounds Dalek-like. The other well known example, although many fewer people have heard it, is a screen reader, such as Jaws or IBM Home Page reader, used by many people with vision impairments.
These basic speech synthesisers provide a very valuable service to people with vision or speech impairments. However, advances in TTS are being driven by new mass-market applications such as mobile phones, in-car communications and sat-nav.
If you are sitting in a top of the range Mercedes you would not be impressed by a tinny voice telling you to "fasten your seat belt". The voice becomes part of the Mercedes corporate image and needs to be as smooth, unique, and as well mannered as the car. As more information is provided to the driver through speech, so the quality of the speech has to improve. Not only must the words be pronounced correctly in context for example, "please close (cloze) the door", and "you are too close (clos) to the car in front", must be pronounced differently and correctly; but also the intonation of the sentences and words must match the context.
SVOX AG is a Swiss based company that specialises in developing TTS technology. It was founded in 2000 as a spin-off from the Swiss Federal Institute of Technology in Zurich (ETH Zurich). Its products are therefore the culmination of over 15 years of research and development. Switzerland is a multi-lingual country and this is reflected in the whole architecture and philosophy of the products.
Having shown that SVOX can produce high quality voice output the next step is to deal with the greater levels of complexity required in-car. These include:
- How to deal with multiple languages, firstly to be able to speak in the preferred language of the driver; secondly, and the greater challenge, how should directions be given as the car moves across borders; should it say Munich or Munchen, and in either case, how should it be pronounced (as a local (Mun-chen) or as read by the listener (Munch-en)?
- Avoiding misleading the on-board computer, which is typically multi-tasking between monitoring the state of the car and other activities (information about a likely engine failure should not be delayed by the TTS engine trying to decipher an SMS).
- Avoiding over-loading the driver (the command "Wake up" should override information that "the weather is going to get worse in half an hour").
- How to add emotion into the voice ("Wake up" should be assertive, while "we are low on fuel but the next services is only 15 kilometres away" should be soothing and calm).
The SVOX TTS engine is scalable between mobile, personal and server solutions so that as these new challenges are solved for the mass market they will become available for the specialised accessibility market. Soon we may hear a Professor Hawking talking emotionally about how happy he is that he can speak with passion, or love, or frustration about his latest discoveries; or even just shout at one of his students "Wake up to the possibilities of TTS!"
Copyright © 2006, IT-Analysis.com