Evaluation of vietnamese speech recognition platforms (vais, viettel, zalo, fpt and google) in news

Abstract

This article introduces an evaluation of Vietnamese Automatic Speech Recognition (VASR) in the news domain from top Vietnamese speech recognition companies such as Vais, Viettel, Zalo, Fpt and top world company such as Google. To evaluate speech recognition systems, Word Error Rate (WER) coefficient with recognized text inputs from Vais VASP, Viettel VASP, Zalo VASR, Fpt VASP and Google VASP platforms were utilized. The recognized texts were acquired by using audio files in the news domain and APIs from Vais VASP, Viettel VASP, Zalo VASR, Fpt VASP and Google VASP platforms to convert from speech to text. The evaluation results obtained from WER which was applied for Vais, Viettel, Zalo, Fpt and Google, show that VASP from Viettel, Zalo, FPT and Google are adequate in which Vais is superior.