Researchers at Microsoft had reported that they have introduced a new speech recognition technology that according to them “transcribes” conversational speech the same way a human does. The error rate of the new Speech Recognition System by Microsoft has said to be decreased to around 5.9 percent.
According to the researchers of Microsoft Technologies their Speech Recognition System has reached, “Human Parity”, which means their speech recognition system is now capable of doing error free translation of human words into its own programming language so as to understand it better and give results the same way. This was also stated by Xuedong Huang who is the Chief Speech Scientist at Microsoft Technologies. He also stated that “This is a historic achievement.”
And it surely is a historical achievement for the researchers as according to them the new speech recognition system is capable of doing almost error free working with the error rate of just 5.9 percent which is near to the work the professional transcriptionists do when they are asked to transcript the speech taken from the same corpus of conversations.
They explained how it work and according to them the new speech recognition system tends to use neural language models that groups similar looking words together in a group, allowing for an efficient generalization of the words.
Microsoft plans to apply this new speech recognition system in their virtual assistant, Cortana which is present in Microsoft devices like Windows 10 and Xbox One.
Microsoft may have achieved something real big according to the present scenario but this doesn’t make it the ultimate king of speech recognition as it needs to make it much powerful that it understands various real life situations too and not just some general terms, also it needs to understand various voice levels that could be from very heavy to very low pitched voice.