This paper focuses on the analysis of spectral phonetic image representations for the Yuhmu language of Ixtenco, Tlaxcala, a tonal language with limited resources, using Implicit Phonetic Segmentation (IPS). The main objective is to develop tools for automatic phoneme segmentation and pronunciation analysis, supporting the study and preservation of this indigenous language. Natural Language Processing (NLP) techniques are employed to analyze the phonetic features of Yuhmu. A total of 66,559 phonetic representations were generated from words in a base dictionary, revealing variability in the representation of phonemes. The obtained results indicated a significant Segment Error Rate (SER) with improvements in segmentation performance through specific parameter combinations, leading to a substantial reduction in SER. In some cases, SER reached optimal levels. Additionally, the average values for MSE, SAM, and SCC demonstrated strong consistency in the analyzed data.

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Phonetic Spectral Image Representation for Yuhmu Language Analysis

  • Eric Ramos-Aguilar,
  • J. Arturo Olvera-López,
  • Ivan Olmos-Pineda

摘要

This paper focuses on the analysis of spectral phonetic image representations for the Yuhmu language of Ixtenco, Tlaxcala, a tonal language with limited resources, using Implicit Phonetic Segmentation (IPS). The main objective is to develop tools for automatic phoneme segmentation and pronunciation analysis, supporting the study and preservation of this indigenous language. Natural Language Processing (NLP) techniques are employed to analyze the phonetic features of Yuhmu. A total of 66,559 phonetic representations were generated from words in a base dictionary, revealing variability in the representation of phonemes. The obtained results indicated a significant Segment Error Rate (SER) with improvements in segmentation performance through specific parameter combinations, leading to a substantial reduction in SER. In some cases, SER reached optimal levels. Additionally, the average values for MSE, SAM, and SCC demonstrated strong consistency in the analyzed data.