Automated rapid prosody transcription
Presentation Type
Abstract
Faculty Advisor
Jonathan Howell
Access Type
Event
Start Date
25-4-2025 1:30 PM
End Date
25-4-2025 2:29 PM
Description
Annotating data, especially speech data, can be a slow and demanding task. To help speed things up, we created a tool called Automated Rapid Prosody Transcription (AutoRPT). This tool helps automatically label sections of speech recordings that relate to tone, rhythm, and emphasis (also known as Prosody) using a method called Rapid Prosody Transcription (RPT). RPT was originally designed to be done by groups of people, getting multiple opinions for each piece of data. AutoRPT works by combining the opinions of two machine learning models that "vote" on how the speech should be labeled. These automatic labels can then be checked and adjusted by human annotators. The goal is to reduce the amount of work humans have to do and make the process faster and easier overall.
Automated rapid prosody transcription
Annotating data, especially speech data, can be a slow and demanding task. To help speed things up, we created a tool called Automated Rapid Prosody Transcription (AutoRPT). This tool helps automatically label sections of speech recordings that relate to tone, rhythm, and emphasis (also known as Prosody) using a method called Rapid Prosody Transcription (RPT). RPT was originally designed to be done by groups of people, getting multiple opinions for each piece of data. AutoRPT works by combining the opinions of two machine learning models that "vote" on how the speech should be labeled. These automatic labels can then be checked and adjusted by human annotators. The goal is to reduce the amount of work humans have to do and make the process faster and easier overall.
Comments
Poster presentation at the 2025 Student Research Symposium.