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 (E2E) models have become the default choice for state-of-the-artspeech recognition systems. Such models are trained on large amounts oflabelled data, which are often not available for low-resource languages.Techniques such as self-supervised learning and transfer learning hold promise,but have not yet been effective in training accurate models. On the other hand,collecting labelled datasets on a diverse set of domains and speakers is veryexpensive. In this work, we demonstrate an inexpensive and effectivealternative to these approaches by ``mining'' text and audio pairs for Indianlanguages from public sources, specifically from the public archives of AllIndia Radio. As a key component, we adapt the Needleman-Wunsch algorithm toalign sentences with corresponding audio segments given a long audio and a PDFof its transcript, while being robust to errors due to OCR, extraneous text,and non-transcribed speech. We thus create Shrutilipi, a dataset which containsover 6,400 hours of labelled audio across 12 Indian languages totalling to4.95M sentences. On average, Shrutilipi results in a 2.3x increase overpublicly available labelled data. We establish the quality of Shrutilipi with21 human evaluators across the 12 languages. We also establish the diversity ofShrutilipi in terms of represented regions, speakers, and mentioned namedentities. Significantly, we show that adding Shrutilipi to the training set ofWav2Vec models leads to an average decrease in WER of 5.8\% for 7 languages onthe IndicSUPERB benchmark. For Hindi, which has the most benchmarks (7), theaverage WER falls from 18.8% to 13.5%. This improvement extends to efficientmodels: We show a 2.3% drop in WER for a Conformer model (10x smaller thanWav2Vec). Finally, we demonstrate the diversity of Shrutilipi by showing thatthe model trained with it is more robust to noisy input.

Effectiveness of Mining Audio and Text Pairs from Public Data for Improving ASR Systems for Low-Resource Languages