In spoken language, a significant proportion of words are produced with missing or underspecified phonetic forms, a process known as phonetic reduction. We have established two different types of reduction for studying reductions in spontaneous speech: lexicalized and nonlexicalized reductions. Lexicalized reductions are regularly encountered in daily language, produced by nearly all speakers, and are related to the lexicon while the non lexicalized reductions are not driven only by the lexicon, occurring at specific points of speech withoutregular reduction patterns. Instead, this type of reduction is influenced by various linguistic and conversational factors. Previously existing methods are not sufficient in detecting all potential reduction s produced in the speech flow. In this study, we propose a novel bottomup approach for identifying reduction prone areas in acorpus of conversations in French, in addition to a manual top down detection based on the known reduced forms. Our novel approach involves the development of a script that automatically detects sequences containing at least 6 phonemes in a 230ms window indicating temporal reduction, allowing the detection of both the lexicalized and non lexicalized sequences.Manually verified extracted sequences were analyzed for the distribution and frequency of reductions, speaker characteristics, as well as the words and phonemes constituting these reduced sequences. Overall, our findings revealed significant variability in terms of reduction frequency and speaker performance, depending on the reduction type, thereby validating thedistinction we established between the two types of reductions. Furthermore, the characteristics of the words and the phonemes present in these sequences were found to influence their likelihood of being reduced.