A Novel Approach for Generating Text Summary of Three Participants Spoken Audio

EasyChair Preprint 3165

Abstract

Data Communication plays a vital role in transferring the information among the people. Auditory is a channel to communicate with each other. The aim of the paper is on computing framework for audio recognition which is further transferred to information and make it available wherever it is necessary by summarizing it. The sections used in this framework are classified into two modules called Training and Testing. Feature extraction of audio is performed under Training phase which further stored in the database. The Testing phase performs the features matching and classification producing the text description of the input audio file. The concept of topic modeling is applied on the text generated and its summary is derived as an output. The database stored for the system is of 100 audio files. The framework establishes a good technique of retrieving the text and generating its summary. The average accuracy rate of the designed system is about 90.06 % with three participant’s consideration.

Keyphrases: Automatic Speech Recognition, Latent Dirichlet Allocation(LDA), Mel Frequency Cepstral Coefficient, Support Vector Machine, Text Summarization, Topic Modeling Procedure

@booklet{EasyChair:3165,
  author    = {Ramesh Kagalkar and Basavaraj Hunshal},
  title     = {A Novel Approach for Generating Text Summary of Three Participants Spoken Audio},
  howpublished = {EasyChair Preprint 3165},
  year      = {EasyChair, 2020}}