28 December 2017 – Computer Science PhD student Chitralekha Gupta won the Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) 2017 Best Student Paper Award.
The APSIPA ASC conference, held at Kuala Lumpur from 12 to 15 December this year, is a gathering of researchers in the field of signal and information, to submit their work for review. A total of 349 accepted papers were published and presented in the conference.
Gupta’s winning paper, “Perceptual Evaluation of Singing Quality (PESnQ)”, developed a perceptually relevant score, PESnQ, that automatically assesses singing quality. With singing pedagogy heavily relying on human evaluation, evaluation criterion for singing is not conveniently available to ordinary people who want to learn the skill.
“PESnQ could serve as a complement to singing lessons, and make singing training more reachable to the masses,” said Gupta.
Gupta authored the paper with NUS Computing Associate Professor Wang Ye and NUS Faculty of Engineering Professor Li Haizhou. To create their new model, the team had to compute features based on singing parameters, like intonation and rhythm, with a cognitive modelling theory used in the Perceptual Evaluation of Speech Quality (PESQ), a well-known speech telecommunication standard.
“The idea of developing a perceptually-relevant standard for singing quality, with parallels to PESQ, the standard in speech telecommunication, was different and appealing,” said Gupta.
To validate their system, the team correlated their proposed singing quality score with ones given by expert human judges, and compared the results with known baseline systems. Results showed that PESnQ’s correlation with human ratings were over 96% more accurate than baseline systems.
“Validating the system was the most challenging part. We had to search for trained musicians who were ready to subjectively judge all the singers in our dataset. Also, this work was highly dependent on singing data that we collected from NUS students, as well as the human validation data for which I am thankful to the trained musicians who agreed to help us out,” Gupta added.
“It was a very pleasant surprise for me [when I learnt of the results], especially given that I had seen some really good work in the conference. It is encouraging to know that my PhD work is indeed headed in the right direction.”
Gupta is in the NUS Graduate School for Integrative Sciences and Engineering, associated with the School of Computing's Computer Science department. She is in the Sound and Music Computing Lab and under the supervision of A/P Wang. Her research area focuses on objective assessment techniques for speech and singing.