A set of Phonological Rules for Mexican Spanish Esmeralda Uraga, Luis Alberto Pineda Cortés In this paper a set of phonological rules for grapheme to phones conversion for Mexican Spanish is presented. These rules permit the automatic generation of phonetic transcriptions and phonetic pronunciation models out of text corpora. A phonetic alphabet called Mexbet, which is based on Worldbet, including repre- sentations for all Mexican phones is proposed. The phonological rules cover most phonetic types of contexts of Mexican Spanish and map orthographic rep- resentations into sequences of Mexbet symbols. The rules have been used for the creation of speech corpora for training acoustic-phonetic models using neural networks, hidden Markov models and hybrid approaches. The speech recogni- tion results and a simple application are also presented. 1 EXTENDED ABSTRACT An essential component of a speech recognition system is the acoustic model. Acoustic models have been created using Neural Networks, Hidden Markov Models or through hybrid approaches. All these models are produced out of a speech corpus by automatic training techniques in which speech signal seg- ments, represented through a number of parameters, are correlated with their corresponding linguistic units, like phonemes or phones. For the creation of reliable and robust models large amounts of data are required. Traditionally phonetic units associated to speech segments are labeled by hand by human ex- perts; however, this is an expensive and time consuming task. In particular, there are no available labeled corpora for Mexican Spanish with the size and quality that the task demands, and the human and material resources to create such cor- pora by hand are not forthcoming. For this reason it is highly desirable to obtain speech data out of text, in the same way that human speakers are able to tell which sounds correspond to sequences of characters when reading. A facility to interpret text and produce the corresponding phonetic units is useful to train the acoustic-phonetic models, diminishing considerably the human effort invested in the task. The rules to map textual representation (graphemes) to phonetic sym- bols underlying such facility will be referred to as phonological rules. The main of purpose of this paper is to present a set of phonological rules for Mexican Spanish, facilitating greatly the creation of acoustic models for this language.