PPA Innovation is responsible for this latest development in artificial intelligence. The local company has enhanced its Thai text-to-speech (TTS) synthesis engine and developed a new version of text-to-speech software that comes with human-like natural voices.

TTS is a speech-synthesis application used to create an audio version of the text in a computer document. It can be used to augment the reading of a text message.

TTS applications include voice-enabled e-mail and spoken prompts in voice-response systems. It is often used with voice-recognition programmes.

President Puttipan Ponyanun said his company spent two years coming up with a natural human-voice text-to-speech system for both Thai and English. It is called PPA Text-to-Speech 2006.

The system was developed using a dynamic algorithm with unit-selection techniques making voices sound human, instead of using the regular Di-Phone technique that makes a robotic-sounding voice.

"With such a natural human-like voice, people won't realise they're talking to a computer," he said.

To develop PPA Text-to-Speech 2006, word-pronunciation and spelling methodologies were analysed using patterns collected in a database of more than 100,000 words.

It is also necessary to know word and sentence segmentations in Thai and English styles.

The system also needs to work out which words are numerical characters and which are alphabetical characters.

"With this logic and methodology, instead of remembering only words in a database, it can always learn new words," said Puttipan.

With these capabilities, the software can be applied widely in real applications, especially call centres and IVR applications.

Normally, messages in IVR systems need to be pre-recorded by humans, meaning that messages cannot be dynamic. Puttipan said that with its software, call centres and messages in IVR systems could be dynamic, with the programme pronouncing words and forming sentences based on people's responses.

"It can do that, but pre-programming is required. It is not necessary to voice record words and sentences in advance. With the dynamic algorithm, the software can read text and pronounce it automatically," said Puttipan.

The development has resulted in two types of products: the server edition and the client edition. The server edition is designed for call and contact centres, and it allows companies to change messages and add new ones as often as they want.

"In the past, this was impossible, because when they wanted to change or add messages, they had to create human voice recordings and put them into the IVR database. This is known as static text, not dynamic text, and it's not convenient or flexible," said Puttipan.

It also can be applied to such emerging applications as news and movie-schedule updates and checking e-mail. The software can create and encourage new companies.

He said the client edition was designed for applications like talking dictionaries and also talking point-of-sale (POS) terminals which will soon be able to "tell" people in supermarkets how much they will have to pay for everything in their cart.

asina@nationgroup.com

Asina Pornwasin

The Nation