Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Acoustic Model Optimization over Multiple Data Sources: Merging and Valuation

Victor Junqiu Wei, Weicheng Wang, Di Jiang, Conghui Tan, Rongzhong Lian

TL;DR

This paper proposes a novel paradigm to solve salient problems plaguing the ASR field, and introduces Shapley Value to estimate the contribution score of the trained models, which is useful for evaluating the effectiveness of the data and providing fair incentives to their curators.

Abstract

Due to the rising awareness of privacy protection and the voluminous scale of speech data, it is becoming infeasible for Automatic Speech Recognition (ASR) system developers to train the acoustic model with complete data as before. For example, the data may be owned by different curators, and it is not allowed to share with others. In this paper, we propose a novel paradigm to solve salient problems plaguing the ASR field. In the first stage, multiple acoustic models are trained based upon different subsets of the complete speech data, while in the second phase, two novel algorithms are utilized to generate a high-quality acoustic model based upon those trained on data subsets. We first propose the Genetic Merge Algorithm (GMA), which is a highly specialized algorithm for optimizing acoustic models but suffers from low efficiency. We further propose the SGD-Based Optimizational Merge Algorithm (SOMA), which effectively alleviates the efficiency bottleneck of GMA and maintains superior model accuracy. Extensive experiments on public data show that the proposed methods can significantly outperform the state-of-the-art. Furthermore, we introduce Shapley Value to estimate the contribution score of the trained models, which is useful for evaluating the effectiveness of the data and providing fair incentives to their curators.

Acoustic Model Optimization over Multiple Data Sources: Merging and Valuation

TL;DR

This paper proposes a novel paradigm to solve salient problems plaguing the ASR field, and introduces Shapley Value to estimate the contribution score of the trained models, which is useful for evaluating the effectiveness of the data and providing fair incentives to their curators.

Abstract

Due to the rising awareness of privacy protection and the voluminous scale of speech data, it is becoming infeasible for Automatic Speech Recognition (ASR) system developers to train the acoustic model with complete data as before. For example, the data may be owned by different curators, and it is not allowed to share with others. In this paper, we propose a novel paradigm to solve salient problems plaguing the ASR field. In the first stage, multiple acoustic models are trained based upon different subsets of the complete speech data, while in the second phase, two novel algorithms are utilized to generate a high-quality acoustic model based upon those trained on data subsets. We first propose the Genetic Merge Algorithm (GMA), which is a highly specialized algorithm for optimizing acoustic models but suffers from low efficiency. We further propose the SGD-Based Optimizational Merge Algorithm (SOMA), which effectively alleviates the efficiency bottleneck of GMA and maintains superior model accuracy. Extensive experiments on public data show that the proposed methods can significantly outperform the state-of-the-art. Furthermore, we introduce Shapley Value to estimate the contribution score of the trained models, which is useful for evaluating the effectiveness of the data and providing fair incentives to their curators.

Paper Structure

This paper contains 20 sections, 2 theorems, 14 equations, 4 figures, 1 table, 3 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Automatic Speech Recognition
  4. Language model
  5. End-to-end ASR model
  6. Acoustic Model Optimization
  7. Incentive Mechanism in Federated Scenario
  8. Problem Setup
  9. Genetic Merge Algorithm
  10. SGD-Based Optimizational Merge Algorithm
  11. Solving Optimization Problem
  12. Discussion on Privacy Issues
  13. Source Model Valuation Based on the Shapley Value
  14. Experiments
  15. Experimental Setup
  16. ...and 5 more sections

Key Result

Proposition 1

$M$ in the form of eq:def_mt covers any model generated by GMA. Besides, term $\Delta W^l$ is brought by the mutation operation. In other words, it always holds that $\Delta W^l=0$ if the mutation operator is not applied.

Figures (4)

  • Figure 1: WERs of the source models and target models generated by different methods.
  • Figure 2: Convergence curves of GMA and SOMA.
  • Figure 3: WERs of the target models optimized on different sizes of the validation set. Log-scale is adopted for $x$-axis.
  • Figure 4: Shapley Value and Average Weight of Source Models

Theorems & Definitions (3)

  • Proposition 1
  • Proof 1
  • Theorem 1: Theorem 3 of jia2019towards