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

An end-to-end (E2E) speaker-attributed automatic speech recognition (SA-ASR)model was proposed recently to jointly perform speaker counting, speechrecognition and speaker identification. The model achieved a lowspeaker-attributed word error rate (SA-WER) for monaural overlapped speechcomprising an unknown number of speakers. However, the E2E modeling approach issusceptible to the mismatch between the training and testing conditions. It hasyet to be investigated whether the E2E SA-ASR model works well for recordingsthat are much longer than samples seen during training. In this work, we firstapply a known decoding technique that was developed to perform single-speakerASR for long-form audio to our E2E SA-ASR task. Then, we propose a novel methodusing a sequence-to-sequence model, called hypothesis stitcher. The model takesmultiple hypotheses obtained from short audio segments that are extracted fromthe original long-form input, and it then outputs a fused single hypothesis. Wepropose several architectural variations of the hypothesis stitcher model andcompare them with the conventional decoding methods. Experiments usingLibriSpeech and LibriCSS corpora show that the proposed method significantlyimproves SA-WER especially for long-form multi-talker recordings.

Hypothesis Stitcher for End-to-End Speaker-attributed ASR on Long-form Multi-talker Recordings