NEC and Nippon Television conducted a demonstration of automatic captioning on Nippon Television’s live broadcast program “Straight News.”
The demonstration used NEC’s DX support service “NEC Enhanced Speech Analysis – high-performance speech analysis -” that utilizes NEC’s AI speech recognition technology. By having the speech recognition model learn broadcasting terms in advance, we confirmed a recognition accuracy of 99.0%.
In addition, They have developed a technology to reduce misrecognition due to ambient noise and BGM, with a view to using it for sports broadcasts, which require noise resistance. NEC’s in-house demonstration confirmed that even in the presence of noise such as train noise and outdoor noise, the recognition accuracy is 88.3%, which is up to 15.9 points higher than speech recognition that is generally used in noisy environments. bottom.
In the broadcasting industry, there is a demand for closed-captioning of television so that necessary information can be accurately conveyed to people with hearing impairments and those who have difficulty hearing TV sounds.
However, in order to realize subtitle broadcasting, manual input by humans is the mainstream, and even when using a speech recognition system, it is necessary to assign multiple proofreaders to ensure the accuracy of the subtitle content. , the problem was that the operational load was large.