Microsoft has been trying to achieve a new level regarding voice technology, and it seems it has managed to make great progresses. The company claims its speech recognition AI reached a new milestone, gaining such an accuracy that it almost challenges humans in terms of recognizing speech and voices.
About a year ago, Microsoft made an announcement regarding the word error rate in speech recognition. There is a standard Switchboard benchmark in the industry which measures this rate, and companies which develop AIs often test it to see how accurate their machines get.
Back then, it reached an impressive word error rate of 6.3 percent. Only a month after the initial announcement, it reached 5.9 percent. If this wasn’t already enough, Microsoft didn’t seem satisfied, and aimed for more progress. The current word error rate is 5.1 percent. This is a huge step ahead, as the company has achieved great progress since last year.
Microsoft reduced its word error rate for voice recognition
Lately, Microsoft has managed to reduce the overall word error rate by 12 percent. This happened mostly due to the improvements it made on the neural networks. For this, they used plenty of linguistic models, together with acoustic samples to help the AI systems get better at recognizing speech.
In other words, Cortana will now be able to better recognize what you’re saying both on PC and on mobile. Also, Microsoft benefits from voice recognition options in Office and in its Cognitive Services as well, and all feature advanced capabilities of picking up words and meanings from speech. Windows also has a quite advanced integrated speech recognition system.
Microsoft has an ultimate goal regarding Cortana. The company is planning to achieve such a great accuracy that users will be able to hold a natural conversation with the AI mechanism and feel like they are talking to a real person. Microsoft is not there yet, but it definitely made some great improvements.
Image Source: Wikimedia Commons