TEVR: Improving Speech Recognition by Token Entropy Variance Reduction

Hajo Nils Krabbenhöft; Erhardt Barth

doi:10.48550/ARXIV.2206.12693

TEVR: Improving Speech Recognition by Token Entropy Variance Reduction

Hajo Nils Krabbenhöft, Erhardt Barth

Abstract

This paper presents TEVR, a speech recognition model designed to minimize the variation in token entropy w.r.t. to the language model. This takes advantage of the fact that if the language model will reliably and accurately predict a token anyway, then the acoustic model doesn't need to be accurate in recognizing it. We train German ASR models with 900 million parameters and show that on CommonVoice German, TEVR scores a very competitive 3.64% word error rate, which outperforms the best reported results by a relative 16.89% reduction in word error rate. We hope that releasing our fully trained speech recognition pipeline to the community will lead to privacy-preserving offline virtual assistants in the future.

Original language	English
Number of pages	10
Specialist publication	arXiv.org
Publisher	arXiv
DOIs	https://doi.org/10.48550/ARXIV.2206.12693
Publication status	Published - 25.06.2022

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

SDG 3 Good Health and Well-being
SDG 4 Quality Education
SDG 9 Industry, Innovation, and Infrastructure
SDG 11 Sustainable Cities and Communities
SDG 12 Responsible Consumption and Production
SDG 14 Life Below Water
SDG 15 Life on Land

Research Areas and Centers

Centers: Center for Artificial Intelligence Luebeck (ZKIL)
Research Area: Intelligent Systems

Access to Document

10.48550/ARXIV.2206.12693

https://arxiv.org/abs/2206.12693

Cite this

@misc{d296d3dcc577472a9a40aeb2dae42432,

title = "TEVR: Improving Speech Recognition by Token Entropy Variance Reduction",

abstract = "This paper presents TEVR, a speech recognition model designed to minimize the variation in token entropy w.r.t. to the language model. This takes advantage of the fact that if the language model will reliably and accurately predict a token anyway, then the acoustic model doesn't need to be accurate in recognizing it. We train German ASR models with 900 million parameters and show that on CommonVoice German, TEVR scores a very competitive 3.64\% word error rate, which outperforms the best reported results by a relative 16.89\% reduction in word error rate. We hope that releasing our fully trained speech recognition pipeline to the community will lead to privacy-preserving offline virtual assistants in the future.",

author = "Krabbenh{\"o}ft, \{Hajo Nils\} and Erhardt Barth",

year = "2022",

month = jun,

day = "25",

doi = "10.48550/ARXIV.2206.12693",

language = "English",

journal = "arXiv.org",

publisher = "arXiv",

}

TY - GEN

T1 - TEVR: Improving Speech Recognition by Token Entropy Variance Reduction

AU - Krabbenhöft, Hajo Nils

AU - Barth, Erhardt

PY - 2022/6/25

Y1 - 2022/6/25

N2 - This paper presents TEVR, a speech recognition model designed to minimize the variation in token entropy w.r.t. to the language model. This takes advantage of the fact that if the language model will reliably and accurately predict a token anyway, then the acoustic model doesn't need to be accurate in recognizing it. We train German ASR models with 900 million parameters and show that on CommonVoice German, TEVR scores a very competitive 3.64% word error rate, which outperforms the best reported results by a relative 16.89% reduction in word error rate. We hope that releasing our fully trained speech recognition pipeline to the community will lead to privacy-preserving offline virtual assistants in the future.

AB - This paper presents TEVR, a speech recognition model designed to minimize the variation in token entropy w.r.t. to the language model. This takes advantage of the fact that if the language model will reliably and accurately predict a token anyway, then the acoustic model doesn't need to be accurate in recognizing it. We train German ASR models with 900 million parameters and show that on CommonVoice German, TEVR scores a very competitive 3.64% word error rate, which outperforms the best reported results by a relative 16.89% reduction in word error rate. We hope that releasing our fully trained speech recognition pipeline to the community will lead to privacy-preserving offline virtual assistants in the future.

U2 - 10.48550/ARXIV.2206.12693

DO - 10.48550/ARXIV.2206.12693

M3 - Article

JO - arXiv.org

JF - arXiv.org

PB - arXiv

ER -

TEVR: Improving Speech Recognition by Token Entropy Variance Reduction

Abstract

UN SDGs

Research Areas and Centers

Access to Document

Fingerprint

Cite this