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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Synthetic data generated by text-to-speech (TTS) systems can be used toimprove automatic speech recognition (ASR) systems in low-resource or domainmismatch tasks. It has been shown that TTS-generated outputs still do not havethe same qualities as real data. In this work we focus on the temporalstructure of synthetic data and its relation to ASR training. By using a noveloracle setup we show how much the degradation of synthetic data quality isinfluenced by duration modeling in non-autoregressive (NAR) TTS. To getreference phoneme durations we use two common alignment methods, a hiddenMarkov Gaussian-mixture model (HMM-GMM) aligner and a neural connectionisttemporal classification (CTC) aligner. Using a simple algorithm based on randomwalks we shift phoneme duration distributions of the TTS system closer to realdurations, resulting in an improvement of an ASR system using synthetic data ina semi-supervised setting.

On the Relevance of Phoneme Duration Variability of Synthesized Training Data for Automatic Speech Recognition