•  
  •  
 

1st Student's Major

Electrical and Computer Engineering and Technology

1st Student's College

Science, Engineering and Technology

Students' Professional Biography

I am an alumnus graduated with a double degree in electrical and computer engineering at Minnesota State University, Mankato. I am originally from Seoul, South Korea. My current goal is to gain work experience for a couple years and then study future for Master's Degree in Computer Science.

Mentor's Name

Rebecca Bates

Mentor's Email Address

rebecca.bates@mnsu.edu

Mentor's Department

Integrated Engineering

Mentor's College

Science, Engineering and Technology

Abstract

High quality automatic speech recognition (ASR) depends on the context of the speech. Cleanly recorded speech has better results than speech recorded over telephone lines. In telephone speech, the signal is band-pass filtered which limits frequencies available for computation. Consequently, the transmitted speech signal may be distorted by noise, causing higher word error rates (WER). The main goal of this research project is to examine approaches to improve recognition of telephone speech while maintaining or improving results for clean speech in mixed telephone-clean speech recordings, by reducing mismatches between the test data and the available models. The test data includes recorded interviews where the interviewer was near the hand-held, single-channel recorder and the interviewee was on a speaker phone with the speaker near the recorder. Available resources include the Eesen offline transcriber and two acoustic models based on clean training data or telephone training data (Switchboard). The Eesen offline transcriber is on a virtual machine available through the Speech Recognition Virtual Kitchen and uses an approach based on a deep recurrent neural network acoustic model and a weighted finite state transducer decoder to transcribe audio into text. This project addresses the problem of high WER that comes when telephone speech is tested on cleanly-trained models by 1) replacing the clean model with a telephone model and 2) analyzing and addressing errors through data cleaning, correcting audio segmentation, and adding words to the dictionary. These approaches reduced the overall WER. This paper includes an overview of the transcriber, acoustic models, and the methods used to improve speech recognition, as well as results of transcription performance. We expect these approaches to reduce the WER on the telephone speech. Future work includes applying a variety of filters to the speech signal could reduce both additive and convolutional noise resulting from the telephone channel.

Creative Commons License

Creative Commons Attribution-No Derivative Works 4.0 International License
This work is licensed under a Creative Commons Attribution-No Derivative Works 4.0 International License.

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.