created a mapping of transcribed audio to the expected output
stored possible uncommon pronunciations of words and their corrected versions in a JSON file
used fuzzy matching to find the best-corrected match for a phrase
filtered out stop words since they were not considered important
stt_mapper returns the corrected output
Some thoughts:
Stop words can be loaded into a hash table.
More possible uncommon pronunciations need to be added to improve the post-editing process. (with a good microphone)
Some thoughts: Stop words can be loaded into a hash table. More possible uncommon pronunciations need to be added to improve the post-editing process. (with a good microphone)