English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

End-to-end automatic speech recognition (ASR) models have seen revolutionaryquality gains with the recent development of large-scale universal speechmodels (USM). However, deploying these massive USMs is extremely expensive dueto the enormous memory usage and computational cost. Therefore, modelcompression is an important research topic to fit USM-based ASR under budget inreal-world scenarios. In this study, we propose a USM fine-tuning approach forASR, with a low-bit quantization and N:M structured sparsity aware paradigm onthe model weights, reducing the model complexity from parameter precision andmatrix topology perspectives. We conducted extensive experiments with a2-billion parameter USM on a large-scale voice search dataset to evaluate ourproposed method. A series of ablation studies validate the effectiveness of upto int4 quantization and 2:4 sparsity. However, a single compression techniquefails to recover the performance well under extreme setups including int2quantization and 1:4 sparsity. By contrast, our proposed method can compressthe model to have 9.4% of the size, at the cost of only 7.3% relative worderror rate (WER) regressions. We also provided in-depth analyses on the resultsand discussions on the limitations and potential solutions, which would bevaluable for future studies.

USM-Lite: Quantization and Sparsity Aware Fine-tuning for Speech Recognition with Universal Speech Models