Paper
Application-Agnostic Language Modeling for On-Device ASR
On-device automatic speech recognition systems face several challenges compared to server-based systems. They have to meet stricter constraints in terms of speed, disk size and memory while maintaining the same accuracy. Often they have to serve several applications with different distributions at once, such as communicating with a virtual assistant and speech-to-text. The simplest solution to serve multiple applicat · Authors: Markus Nußbaum-Thom, Lyan Verwimp, Youssef Oualil · Published 2023-05-16 · arXiv:2305.09764
- Published
- 2023-05-16
- Authors
- 3
- arXiv
- 2305.09764
- Category
- cs.CL
Abstract
arXiv:2305.09764
Markus Nußbaum-Thom, Lyan Verwimp, Youssef Oualil
On-device automatic speech recognition systems face several challenges compared to server-based systems. They have to meet stricter constraints in terms of speed, disk size and memory while maintaining the same accuracy. Often they have to serve several applications with different distributions at once, such as communicating with a virtual assistant and speech-to-text. The simplest solution to serve multiple applications is to build application-specific (language) models, but this leads to an increase in memory. Therefore, we explore different data- and architecture-driven language modeling approaches to build a single application-agnostic model. We propose two novel feed-forward architectures that find an optimal trade off between different on-device constraints. In comparison to the application-specific solution, one of our novel approaches reduces the disk size by half, while maintaining speed and accuracy of the original model.
Related papers
14
-
University of California, Berkeley
-
University of California, Santa Cruz
-
Tribhuvan University
-
University of Texas at Arlington
-
Harvard University
-
The University of Texas at Austin