“Many challenges to localize services in Indonesian,” he said. First, Indonesian is homograph writing the same word but different pronunciation and meaning as the word apple (ceremony) and apple (fruit), voice search should be able to distinguish.
Second, Indonesian and even after there is EYD (spelling enhanced) still affected many Dutch words. For example, the name of Sukarno and Soekarno and the same pronunciation but different writings. Voice search should be able to overcome it.
Third, some people use the phrase mixture Indonesia Indonesian and English. Services must be able to recognize words in different languages in one sentence.
Fourth, there is currently no standardization of the pronunciation of words in Indonesian. EYD new to writing, but to the pronunciation depends on access so that pronunciations vary depending on ethnic origin or region of origin of speakers.
“Google since last year to recruit and provide time for the engineer to the data collected on,” said Heng. He said one way is to involve as many as 500 student volunteers. They were asked to pronounce the word list so that the collected database to hundreds of thousands of phrases. This database is then stored in a server cloud that will translate the search text.
According to the story, Heng who is directly involved in the project, he also sent to one of Google’s research center in New York to improve the model’s voice search in Indonesian language. Machine was tried to do a search with voice over and over again until the level of truth as possible. And over time, machine learning new phrases from many users of the service.
Related posts:
- Google Machine Continues Learning Indonesian-1 As a technology company, one of the serious works with...
- Does Google Penguin Hit Your Site? Google in their official webmaster blog promise that they will...