Network-based Speech Translator

Updated Researches

sst The language barrier is a major problem for today's global communication. Existing technologies like automatic text translation as well as speech input/output can now be a seed for an innovative technology called speech-to-speech translation, where speech in a source language can be translated to speech in a target language.

Under the collaboration among 8 Asian countries; Thailand, Japan, Korea, Vietnam, Indonesia, China, India, and Singapore, in the name “Asian Speech Translation Advance Research Consortium (A-STAR)”, a network-based speech translation system has been developed. The system facilitates people to talk to each other by using their own mother-tongue language. Member countries set up basic speech translation components including automatic speech recognition (ASR) and text-to-speech synthesis (TTS) for their own languages, and statistical machine translation (SMT) between their languages to English or others. The prototype Internet-based engine links web services of such basic components, forming a speech translator for 8 languages plus English.

asia_map

fig4_serviceprocedure

Client applications can be in many forms such as voice-interfaced instant messengers with a translation service, and a peer-to-peer software that helps translation between two languages. Although the current system has been built particularly for the travel domain, modification to other domains requires only the development of a domain-specific database. This happens to be useful in many application areas e.g. tourism, multi-cultural communication, social networking, international conference and business.

Key Features

Speech translation among 8 Asian languages (Thai, Japanese, Korean, Indonesian, Vietnamese, Chinese, Malay, and Hindi) and English
For the Thai-English pair developed by NECTEC, the system recognizes 14,000 Thai words and 21,000 English words in the travel domain. Users are allowed to speak any sentence based on recognizable words.
Ubiquitous computing via the Internet interface
Less than 5 seconds/transaction on the normal Internet condition
Overall translation accuracy varies between 60-90% depending on the speaking environment and style
Domain, environment, and speaker adaptability and scalability

Contact: Human Language Technology Laboratory National Electronics and Computer Technology Center
http://www.hlt.nectec.or.th/

R & D

Highlight Products

Magneto Transistor

This device inspects or measures the magnetic field. It is produced with the semiconductor technology and is designed to have the uncomplicated structure of transistor.

ISFET

This product is newly developed to substitute the previous glass bulb sensor. This sensor is made from silicon and designed to have the structure of ISFET.

CMOS

This type of products includes all circuits that are made from CMOS silicon. This kind of circuit has been widely used to the extent where we can say that 90 percent of the manufactured semiconductor circuits in the industry are appliances or circuits that consist of CMOS silicon.