In the paper, we studied and implemented the system for CALL systems based on the NLP technologies-bilingual system for spoken Chinese and Hakka with four-county accent. The online system provides mainly several tasks of learning; bilingual lexical information, transliteration of words and examples, specially the function for practice of oral speeches. The bilingual corpora, dictionary and speech database are generated. In our speech databases, two kinds of unit, character and word speeches units for Hakka are recorded. There are 1408 monosyllabic speech units for Chinese, 2500 HAKKA's characters and 2200 word speeches for synthesis unit, respectively. The bilingual lexicon is bidirectional and allows the translation to and from both languages. The synthesized examples of speeches were evaluated for the quality of intelligibility and the results of satisfaction testing for two kinds of units achieve 5.31 and 7.82, respectively. It is obvious that our bilingual system provides real speech with natural quality and helpful lexical information on web, and achieve the goals for language learning.