Accession Number:

AD1038283

Title:

Multilingual Data Selection for Low Resource Speech Recognition

Descriptive Note:

Conference Paper

Corporate Author:

IBM THOMAS J WATSON RESEARCH CENTER YORKTOWN HEIGHTS NY YORKTOWN HEIGHTS United States

Personal Author(s):

• Thomas, Samuel
• Audhkhasi,Kartik
• Cui,Jia
• Kingsbury,Brian
• Ramabhadran,Bhuvana

Report Date:

2016-09-12

Pagination or Media Count:

5.0

Abstract:

Feature representations extracted from deep neural network-based multilingual frontends provide significant improvements to speech recognition systems in low resource settings. To effectively train these frontends, we introduce a data selection technique that discovers language groups from an available set of training languages. This data selection method reduces the required amount of training data and training time by approximately 40, with minimal performance degradation. We present speech recognition results on 7 very limited language pack VLLP languages from the second option period of the IARPA Babel program using multilingual features trained on up to 10 languages. The proposed multilingual features provide up to 15 relative improvement over baseline acoustic features on the VLLP languages.

Descriptors:

• artificial neural networks
• artificial intelligence
• digital data
• automated speech recognition
• recognition

Subject Categories:

• Cybernetics
• Voice Communications

Distribution Statement:

APPROVED FOR PUBLIC RELEASE