Automatic Phonetic Segmentation of the Yuhmu Language Using Mel Scale Spectral Parameters

The application of digital signal processing techniques and machine learning, along with implicit segmentation, poses a challenge in the study of phonetic segmentation of indigenous languages in Mexico, given their linguistic and phonetic diversity. The analysis of Mel-scaled spectrograms offers an effective approach to identify patterns that can outline relevant information. By comparing the results with the actual number of phonemes in a word, both successes and areas for improvement can be observed. This article proposes a methodology for automatic segmental analysis of the Yuhmu language, considering parameter search in the Mel scale and implementing the cosine distance between spectrogram vectors. Additionally, relevant data within the resulting matrices are taken into account based on four key thresholds in information selection. The analysis yields a Segment Error Rate (SER) ranging from 38.79% to 41.35%, which aligns with the results reported in the literature on the subject.