Earticle

초록

영어

The identification and association of phonemes from different shapes of the vocal tract can be used for multiple tasks such as diagnosing pronunciation difficulties in patients through the analysis of images like the MRI. However, the lack of reliable vocal tract datasets makes tasks like those hard to be accomplished. Through this paper, an initial proposal on how to make a vocal tract dataset is made and how it could be potentially applied for classifying phonemes. For the creation of the dataset the Vocal Tract Lab Python API was utilized, and those generated images were used as input for training the classifier. The vocal tract images were made from different ages and genders. Only phonemes representing vowels are analyzed and the quantity of the images created for the training are small, which made the test results from the phoneme classification fluctuate in each training run. Still, the current work represents an initial step towards new works in this direction.

목차

Abstract
1. Introduction
2. Methods
2.1. Dataset
2.2. Experimental setup
3. Experimental result
4. Conclusions
Acknowledgement
References

저자

• Rodrigo Picinini Méxas [ Department of Computer Science and Engineering, Sogang University ]
• Unsang Park [ Department of Computer Science and Engineering, Sogang University ] Correspondence author

참고문헌