ASRU 2021 : IEEE Automatic Speech Recognition and Understanding Workshop

Top Research Topics at IEEE Automatic Speech Recognition and Understanding Workshop?

• Speech recognition (97.71%)
• Artificial intelligence (71.94%)
• Natural language processing (35.59%)

The topics of Speech recognition, Artificial intelligence, Natural language processing, Pattern recognition and Speech processing are the focal point of discussions in the conference. IEEE Automatic Speech Recognition and Understanding Workshop dives deep in exploring the relationship between the study of Speech recognition and Natural language. While work presented in the conference provided substantial information on Artificial intelligence, it also covered topics in Machine learning and Vocabulary.

The research on Natural language processing featured in IEEE Automatic Speech Recognition and Understanding Workshop combines topics in other fields like Pronunciation and Speech corpus, Speech analytics. Some problems in Pattern recognition that were presented in the conference overlapped with concepts under Artificial neural network, Feature (machine learning) and Robustness (computer science). Speech processing research is concerned with Voice activity detection in particular.

Aside from investigating topics in Perplexity under Language model, the conference also explores concepts in Cache language model. Discussions in IEEE Automatic Speech Recognition and Understanding Workshop are anchored in the subject of Word error rate and the similar topic of Word (computer architecture). IEEE Automatic Speech Recognition and Understanding Workshop explores research in Hidden Markov model and the adjacent study of Context (language use).

What are the most cited papers published at the conference?

• The Kaldi Speech Recognition Toolkit (4164 citations)
• A post-processing system to yield reduced word error rates: Recognizer Output Voting Error Reduction (ROVER) (1129 citations)
• Hybrid speech recognition with Deep Bidirectional LSTM (1091 citations)

Research areas of the most cited articles at IEEE Automatic Speech Recognition and Understanding Workshop:

The most cited publications explore disciplines such as Speech recognition, Artificial intelligence, Natural language processing, Word error rate and Hidden Markov model. The most cited papers investigate Speech recognition research which frequently intersects with Artificial neural network. The most cited papers explore topics in Artificial intelligence which can be helpful for research in disciplines like Machine learning, Vocabulary and Pattern recognition.

What topics the last edition of the conference is best known for?

• Artificial intelligence
• Speech recognition
• Machine learning

The previous edition focused in particular on these issues:

The conference was organized to reinforce research efforts on Speech recognition, Artificial intelligence, Artificial neural network, Pattern recognition and Word error rate. In addition to Speech recognition research, the event aims to explore topics under Time delay neural network and Recurrent neural network. The work on Artificial intelligence tackled in the conference brings together disciplines like Context (language use), Machine learning and Natural language processing.

Issues in Natural language processing were discussed, taking into consideration concepts from other disciplines like Speech corpus and Word (computer architecture). IEEE Automatic Speech Recognition and Understanding Workshop focuses on Artificial neural network but the discussions also offer insight into other areas such as Latent Dirichlet allocation, Speaker diarisation, Conditional random field, Mixture model and TIMIT. While the event focused on Pattern recognition, it was also able to explore topics like Feature (machine learning) and Autoencoder.

The most cited articles from the last conference are:

• The third ‘CHiME’ speech separation and recognition challenge: Dataset, task and baselines (463 citations)
• EESEN: End-to-end speech recognition using deep RNN models and WFST-based decoding (392 citations)
• Applying deep learning to answer selection: A study and an open task (233 citations)

Papers citation over time

A key indicator for each conference is its effectiveness in reaching other researchers with the papers published at that venue.

The chart below presents the interquartile range (first quartile 25%, median 50% and third quartile 75%) of the number of citations of articles over time.

Top authors and change over time

The top authors publishing at IEEE Automatic Speech Recognition and Understanding Workshop (based on the number of publications) are:

• Mark J. F. Gales (22 papers) published 8 papers at the last edition, 7 more than at the previous edition,
• Satoshi Nakamura (21 papers) published 5 papers at the last edition, 4 more than at the previous edition,
• Hermann Ney (20 papers) published 2 papers at the last edition, 1 more than at the previous edition,
• Lin-Shan Lee (20 papers) published 3 papers at the last edition, 2 more than at the previous edition,
• Bhuvana Ramabhadran (18 papers) published 1 paper at the last edition, 4 less than at the previous edition.

The overall trend for top authors publishing at this conference is outlined below. The chart shows the number of publications at each edition of the conference for top authors.

Top affiliations and change over time

Only papers with recognized affiliations are considered

The top affiliations publishing at IEEE Automatic Speech Recognition and Understanding Workshop (based on the number of publications) are:

• IBM (63 papers) published 2 papers at the last edition, 7 less than at the previous edition,
• Microsoft (44 papers) published 7 papers at the last edition, 5 more than at the previous edition,
• Carnegie Mellon University (36 papers) published 4 papers at the last edition, 6 less than at the previous edition,
• University of Cambridge (36 papers) published 10 papers at the last edition, 8 more than at the previous edition,
• Johns Hopkins University (27 papers) published 4 papers at the last edition, 2 more than at the previous edition.

The overall trend for top affiliations publishing at this conference is outlined below. The chart shows the number of publications at each edition of the conference for top affiliations.

Publication chance based on affiliation

The publication chance index shows the ratio of articles published by the best research institutions at the conference edition to all articles published within that conference. The best research institutions were selected based on the largest number of articles published during all editions of the conference.

The chart below presents the percentage ratio of articles from top institutions (based on their ranking of total papers).Top affiliations were grouped by their rank into the following tiers: top 1-10, top 11-20, top 21-50, and top 51+. Only articles with a recognized affiliation are considered.

During the most recent 2015 edition, 3.74% of publications had an unrecognized affiliation. Out of the publications with recognized affiliations, 30.10% were posted by at least one author from the top 10 institutions publishing at the conference. Another 12.62% included authors affiliated with research institutions from the top 11-20 affiliations. Institutions from the 21-50 range included 30.10% of all publications and 27.18% were from other institutions.

Returning Authors Index

A very common phenomenon observed among researchers publishing scientific articles is the intentional selection of conferences they have already attended in the past. In particular, it is worth analyzing the case when the authors participate in the same conference from year to year.

The Returning Authors Index presented below illustrates the ratio of authors who participated in both a given as well as the previous edition of the conference in relation to all participants in a given year.

Returning Institution Index

The graph below shows the Returning Institution Index, illustrating the ratio of institutions that participated in both a given and the previous edition of the conference in relation to all affiliations present in a given year.

The experience to innovation index

Our experience to innovation index was created to show a cross-section of the experience level of authors publishing at a conference. The index includes the authors publishing at the last edition of a conference, grouped by total number of publications throughout their academic career (P) and the total number of citations of these publications ever received (C).

The group intervals were selected empirically to best show the diversity of the authors' experiences, their labels were selected as a convenience, not as judgment. The authors were divided into the following groups:

• Novice - P < 5 or C < 25 (the number of publications less than 5 or the number of citations less than 25),
• Competent - P < 10 or C < 100 (the number of publications less than 10 or the number of citations less than 100),
• Experienced - P < 25 or C < 625 (the number of publications less than 25 or the number of citations less than 625),
• Master - P < 50 or C < 2500 (the number of publications less than 50 or the number of citations less than 2500),
• Star - P ≥ 50 and C ≥ 2500 (both the number of publications greater than 50 and the number of citations greater than 2500).

The chart below illustrates experience levels of first authors in cases of publications with multiple authors.