F
F
Flatout732017-05-08 18:18:35
Speech recognition
Flatout73, 2017-05-08 18:18:35

What algorithm would you recommend for minimizing errors in speech recognition?

There is original text and text recognized in real time using the bing speech api. In general, both texts should match, but there may be deviations (that is, a person speaks from the original text, but can start from anywhere or interrupt for a while and start talking off topic, and then continue reading through the text). How can I get rid of speech recognition errors? The api itself provides error handling in words, that is, such words exist, but a person could pronounce a completely different word, and not the one that the framework recognized, how can one understand, based on neighboring recognized words, that this is just a recognition error, and not a person began to read from another place? Need to use some kind of fuzzy search algorithm for words in sentences?

Answer the question

In order to leave comments, you need to log in

2 answer(s)
X
xmoonlight, 2017-05-10
@xmoonlight

Try to connect the Stumper.ru API. It gives
very accurate results for matching two strings.
ede39a35933e42edbc9da3408edb748c.jpg

D
Dimonchik, 2017-05-08
@dimonchik2013

see NLTK
bigrams, trigrams and statistics
about "I read from here, but now not from here" it will be more complicated, but in general the same

Didn't find what you were looking for?

Ask your question

Ask a Question

731 491 924 answers to any question