The new milestone of Microsoft indicates that, for the first time, a computer can identify the words in a conversation similar to what a person might do. “Our study group arrived at 5.1% rate of error for our system of speech recognition. This is a new milestone in the industry, considerably exceeding the correctness we gained previous year,” Microsoft claimed late last week in a blog post.
The group from Microsoft Artificial Intelligence and Research last year in October reported a system for speech recognition that makes the similar or less errors as compared to that with the professional transcriptionists. The scientists had then posted a WER (Word Error Rate) of 5.9%.
“In 2016, dialog and speech research team of Microsoft declared a milestone in arriving at human parity on the recognition task of ‘Switchboard’ conversational speech. This indicated that we had made technology that recognized words in professional human transcribers as well as a conversation,” claimed Technical Fellow at Microsoft, Xuedong Huang, to the media in an interview. “Switchboard” is a body of telephone conversations that are recorded. The speech research group has utilized it for over 20 Years to remark systems for speech recognition.
The task comprises transcribing chats between strangers taking about topics such as politics and sports. The group utilized CNTK (Microsoft Cognitive Toolkit 2.1), the most scalable profound learning software obtainable, for discovering sample architectures. In addition to this, investment of Microsoft in infrastructure of cloud compute, particularly in Azure GPUs, assisted to enhance the speed and effectiveness.
Arriving at human parity with correctness similar to humans has been a research objective for the previous 25 Years. “The willingness of Microsoft to spend in long-term study is now paying dividends for our users in services and products such as Presentation Translator, Cortana, and Microsoft Cognitive Services,” the post further read. “Shifting to understanding speech from distinguishing is the upcoming huge frontier for speech tech,” the post further added.
Well, for now, we wish all the luck to Microsoft so that it can progress more in its field and benefit the users with the advancements in the tech sector.