This research concerns the identification of a set of multi-lingual poly-phonemes, which are considered similar enough to be defined across a number of European languages. Results are given of a data-driven approach contrasting to our earlier approach, which was based on auditory phonetic decisions and phonetic statements in the literature. The results show that it is possible to substitute language-specific models with corresponding poly-phonemes trained on other languages and in doing so increase the performance.
Keywords: Phoneme identification, language-independent poly-phonemes, language-dependent mono-phonemes.