We have been working for a few years now in the research and development of 1000 words, isolated word, speaker independent speech recognition in Spanish using discrete HMM. Since we published the last available methods and results, we have been optimizing and improving the system in several directions: Using a new way for phoneme duration incorporation Using a new front end Augmenting the database Using context dependent models and different types of smoothing With all improvements, we have reached 84.18% word recognition for 1006 words vocabulary without grammar.