We add new languages periodically. In the process of adding a new language to the automatic translation system, the most important learning material is existing high-quality translations—the same text in two languages.
A significant amount, typically 4 to 5 million words of this type of text, is needed to build a reasonable-quality automatic translation system for a particular language pair. Find out how automatic translation works at www.aka.ms/MT.
Speech recognition is a totally different process than translation. It requires a whole new set of data (hundred of hours of recorded audios and their associated transcriptions) and trainings of our machine learning based engines to become available. Therefore, we cannot always deliver speech recognition as quickly as new supported languages become available for Microsoft Translator.
We continue to release improvements to the quality of our existing languages and to expand on our coverage of languages from around the world so that more and more people can talk to the Microsoft Translator in their own language.