Speech-to-Text: How to Improve Output Accuracy to make Speech Recognition more Accessible

This paper analyzes the newest algorithms being used in speech-to-text programs in order to produce the most accurate transcripts possible. Metadata extraction is the main focus, as well as the addition of punctuation prediction. Work was tested on both news speech and conversational speech to produce valid comparisons. Modern applications were also tested to help show where we stand today and where we can improve upon in the future. Knowing human tendencies and understanding how human listening skills work are how scientists are approaching the remaining issues today. In conclusion, this paper will discuss possible future routes for further improvements.