Formosa Speech Recognition Challenge Workshop 2018

The FSR-2018 Workshop is a satellite event of ISCSLP 2018

Time: 14:0016:00, November 27, 2018

Venue: Poster Session area, 3F, Humanities and Social Sciences Building, Academia Sinica, Taipei [link]

Program: Single Poster Presentation [link]

Cost: There is no cost to attend the FSR Workshop, thanks to ISCSLP 2018

Call for Participation

The FSR-2018 Workshop is the culmination of the Formosa Speech Recognition Challenge 2018 which is an open Taiwanese Broadcast Mandarin speech recognition evaluation campaign using common test sets. The aims of the workshop are to present the results and for participants in the Challenge to describe their systems.

Who can attend the workshop ?

The workshop is Open to all and Free (no registration fee) and we encourage participation from anyone interested in Taiwanese Mandarin speech recognition.

Who can submit a paper to the workshop ?

All participants in the Challenge are expected to submit a paper describing their entry (even if they cannot attend the workshop in person). Papers will be refereed by the Programme Committee and the Expert Committee.

Paper submission instructions

All submissions should be camera-ready PDF files of up to 5 pages in length. If the paper consists of 5 pages, the last page may only contain references. All papers must conform to the official double-column format in accordance with ISCSLP 2018 paper format requirement.
The working language of FSR-2018 is English.
FSR is a scientific investigation - we are all trying to understand why some techniques work better than others.
With this in mind, please write a detailed, technical paper aimed at a specialist audience. Focus on analysis and evaluation. Try to explain why your system performed the way it did, and what makes it different from other systems. Explain why your system is designed in a particular way. For example, report internal evaluations you have done to select certain methods.
Submit your paper by email to yfliao@mail.ntut.edu.tw before 23:59, 11/25/2018 (Taipei Time, GMT+8) (deadline extended)

Organisers

Yuan-Fu Liao, Taipei University of Technology, Taipei
Hsin-Min Wang, Academia Sinica, Taipei

Practical information

Registration procedure

Please register in advance to help us to better prepare the workshop. Here is the Registration Form [link]. Thank a lot!

Cost

There is no cost to attend the FSR Workshop, thanks to ISCSLP 2018.

Travel

Accommodation

No accommodation is arranged by the workshop: please make your own arrangements.

Presentation Instructions

Poster size

Each Team will have a Magnetic Whiteboard with Width 105 cm * Height 140 cm

Place

Poster Session Area, 3F

Challenge Summary

Participants

There were a total of 26 teams: 8 industrial, 14 academic and 5 individuals.

Survey

Survey of occupation, speech data and lexicon usage, toolkit and approach adopt: the questions are: (a) Are you from industrial sector?, (b) Did you use the provided speech data?, (c) Did you use the provided lexicon?, (d) Is your system built by using the Kaldi Speech Recognition toolkit?, (e) Did you adapt the end-to-end approach? Here the proportions of answer "yes" and "no" are shown and marked as blue and red color, respectively.

Approaches

The resources used and strategies adopt by the participants int the final-test evaluation. Due to space limitation, only the information of top 4 teams (ID: "K", "G", "C" and "R") are shown.

Conclusions

Final-test is more difficult!
Baseline degraded from 17% to 25%. May due to Children’s recordings (elementary school, grade 5~6)
There are in total, 16 teams submitted 30 recognition results.
12 teams beated iFlyTek’s (18.8%) and Goolge’s (20.6%) commercial systems.
The performance of all teams were dramatically improved during the challenge period.

Awards

Best Industrial System: ASUS Inc. (8.1% CER)
Best Academic System: National Taiwan Normal University (9.4% CER)

Proceeding of FSR-2018 Workshop

Hsiu-Jui Chang, Wei-Cheng Chao, Tien-Hong Lo, Berlin Chen, NTNU Speech Recognition System at FSR 2018
Hung-Shin Lee, Kuan-Yu Chen, Yu Tsao, Hsin-Min Wang, The AS Kaldi-based Taiwanese Mandarin ASR System for FSR-2018
Meng-Che Wu, Wei-Yuan Chen, Alim Misbullah, Established a Taiwanese Speech Recognition System for Formosa Speech Recognition Challenge 2018
Hsiao-Tsung Hung, The AlexHT system for FSR Challenge
Hong-Bin Liang , Yih-Ru Wang, The NCTU ASR System for Formosa Speech Recognition Challenge 2018
Meng-Ping Lu and Chia-Ping Chen, NSYSU Team For The Formosa Speech Recognition Challenge 2018
Ming-Han Yang, Yao-Chi Hsu, Yu-Chen Kao, Tian-Ming Xu, Yun-Wen Li, Berlin Chen, The DMS-ASR System for the Formosa Speech Recognition Challenge 2018
Li-Hsuan Chen, Chieh-Kuo Hu, Ling-Ju Hung, Chen-Wan Lin, Towards a Robust Taiwanese Mandarin Automatic Speech Recognition System with Kaldi Toolkit