Predicting the Evolution of Human Speech Patterns Using Machine Learning Models
Can a person predict it using simple methods? If so, how and what the results will look like? I covered these questions using sources such as TED Talks from 2009 to 2016 and Supreme Court transcripts with start and stop times from 1955 to 2019. I created and used my model that predicts how the speed of speech will change based on previously available data. There were four methods by which my model could predict the WPM: copycat - surprisingly, it is the simplest method among all, it takes the most recent data and implies to the data that should be determined; all-history trend - popular method to predict virtually anything; 5-year average, or short ruler, - method with which model creates a visual ruler and puts it at the last five years, pointing where the next data will be; 20-year trend, or long ruler, - similar to short ruler that uses a longer line and is more reliable since more data is included. Four rules were used to prevent results from low quality data to ensure results were applicable. Results show that human speech evolution does not stop, constantly fluctuating. Overall, mathematical and machine learning models can be used to forecast the human speech evolution, and one of the best methods was the “ruler” methods, winning most of the rounds. In conclusion, the WPM is contingent on who speaks and where spoken. For example, the WPM for Supreme Court observed an increase while, unexpectedly, that of TED Talks went down
Authors
- Farruxbek Xolmuminov
Institutions
- Westminster International University in Tashkent (UZ)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-19
- DOI
- https://doi.org/10.5281/zenodo.24360786
- Primary Topic
- Speech Recognition and Synthesis
- Type
- preprint