Hello,
Transcription of any possible word spoken in a language isn’t a feature of offline recognition that is performed on a handheld device, sorry. In offline recognition on a handheld device, you can work with vocabularies of maybe up to 2000 words which you can choose, and you can also switch between them, but you can’t use vocabularies with the hundreds of thousands of words that would be needed in order to understand any possible utterance a user might say, since among other things, that would require more space than the available memory on the device for a single app. You will want to investigate online services providing transcription services, such as the Nuance NDK.