Browse Prior Art Database

System and Method for Dynamic Language Model Adaptations for Different Regional Accents for Name Recognition in a Given Language

IP.com Disclosure Number: IPCOM000244922D
Publication Date: 2016-Jan-29
Document File: 3 page(s) / 149K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is an automated process to improve the results of automatic speech recognition (ASR) in mobile applications when the user speaks English with the accent of another language. The system accepts speech and word input from users with native accents, stores various speech samples of the same English word or noun contributed by different regional users, and instantly or periodically recompiles the language data model used by ASR units.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 53% of the total text.

Page 01 of 3

System and Method for Dynamic Language Model Adaptations for Different Regional Accents for Name Recognition in a Given Language

Many mobile applications (apps) require speech recognition based services. The results of speech recognition services are not always accurate because the services cannot understand the native and regional accents of users speaking English .

Current speech systems for name lookup work very well when using the base language in which the model was produced. If usage deviates from the base language model , then automatic speech recognition (ASR) success significantly declines depending on the language. Some solutions try to improve the speech recognition by switching language models.

A method is needed to improve the recognition by enhancing the one language model .

The novel contribution is an automated process for accepting speech input as well as

word input from users with native accents and storing various speech samples of the same English word or noun contributed by different regional users . The process also instantly or periodically recompiles the language data model used by automated speech recognition units to improve the speech recognition of a system .

The core components of the process are methods to :

1. Accept speech samples and spellings of a spoken word /noun from a non-native speaker


2. Record/store such speech samples from many users who have spoken the same

word in the associated regional accent

3. Instantaneously or periodically recompile the language data model for improved speech recognition

This process and system improv...