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

State-of-the-art automatic speech recognition (ASR) system development isdata and computation intensive. The optimal design of deep neural networks(DNNs) for these systems often require expert knowledge and empiricalevaluation. In this paper, a range of neural architecture search (NAS)techniques are used to automatically learn two types of hyper-parameters offactored time delay neural networks (TDNN-Fs): i) the left and right splicingcontext offsets; and ii) the dimensionality of the bottleneck linear projectionat each hidden layer. These techniques include the differentiable neuralarchitecture search (DARTS) method integrating architecture learning withlattice-free MMI training; Gumbel-Softmax and pipelined DARTS methods reducingthe confusion over candidate architectures and improving the generalization ofarchitecture selection; and Penalized DARTS incorporating resource constraintsto balance the trade-off between performance and system complexity. Parametersharing among TDNN-F architectures allows an efficient search over up to 7^28different systems. Statistically significant word error rate (WER) reductionsof up to 1.2% absolute and relative model size reduction of 31% were obtainedover a state-of-the-art 300-hour Switchboard corpus trained baseline LF-MMITDNN-F system featuring speed perturbation, i-Vector and learning hidden unitcontribution (LHUC) based speaker adaptation as well as RNNLM rescoring.Performance contrasts on the same task against recent end-to-end systemsreported in the literature suggest the best NAS auto-configured system achievesstate-of-the-art WERs of 9.9% and 11.1% on the NIST Hub5' 00 and Rt03s testsets respectively with up to 96% model size reduction. Further analysis usingBayesian learning shows that ...

Neural Architecture Search For LF-MMI Trained Time Delay Neural Networks