The sound of an identity: A look into the Singaporean accent

Methodology: We used the National Speech Corpus (NSC), developed by the Infocomm Media Development Authority, for the main dataset. It was developed in three parts, from approximately 1,000 hours of recordings of phonetically balanced scripts from about 1,000 local English speakers, for each part. In total, more than 3,000 hours of audio recordings from more than 2,000 participants of different ages, genders and local ethnic groups across Singapore were used to create the NSC.

For this project, we looked at only one part of the dataset, which recorded participants reading sentences from local newspapers, blogs and websites, and chose one specific sentence: “They tried to keep it a secret but he was able to figure everything out.” The sentence was read out by 980 respondents. From these recordings, we removed those where the sentence was not read correctly, and we manually tagged the rest of the recordings for the variables mentioned in the piece.

We built a custom variable typeface to encode pronunciation, pitch and duration data of an accent. To visualise the pitch data, we first found the median pitch for each speaker, then calculated how much each word’s pitch deviated from that median. We mapped these values to the vertical position of each word. For duration, we identified the shortest and longest words for each speaker and mapped them to the narrowest and widest word widths. The widths of the remaining words were scaled accordingly.