[ad_1]
Recognize about 99 percent. words
“All the television presenters in the country are true language professionals; in the ten-point system, they would receive 10 points. So when we talk about the top, let’s keep in mind that we are comparing the best with the best. We see that intelligence artificial has slipped in just a few words, it recognized the absolute majority of the content provided by knowledge leaders flawlessly, “said Deividas Butkus, Head of Communication Monitoring and Analysis Department at Kantar.
According to him, the speech of the knowledge leaders in the first six positions was recognized by artificial intelligence in more than 99 percent. and very easily handled the information provided by Marijus Žiedas (LRT) and Lina Kairytė (LNK) (approximately 99.5 precision).
“The country’s knowledge leaders undoubtedly have excellent diction, and artificial intelligence is constantly improving, for example, when we started in 2018, it acknowledged an average of 95.55 percent. speeches of knowledge leaders, and this year – already 98.56. In some cases, presenters are only half a percent short of 100, as in the case of Mary Ring.
And Justė Tamošaitienė (LRT) has reached the best record of all time so far, 99.85 percent. These voices can be used as a standard for Lithuanian speech, ”said expert D.Butkus.
Justė Tamošaitienė and Antanas Tamošaitis / Photo by Lukas Balandis
Instant messaging – LNK
The biggest instant conversations between news anchors are found on the LNK channel. Its knowledge leaders Karolina Liukaitytė, Eglė Šantaraitė, Paulius Skučas and Aurelijus Bezekavičius can read more than 720 words in 5 minutes, maintaining around 98 percent. speech recognition. For example, Lina Kairytė speaks a third less in five minutes than Karolina Liukaitytė, but the recognizability of her speech is among the record, around 99.5 percent. Colleague Gintaras Deksnys (99.38%), who also speaks at a calmer pace, is not much inferior to her.
The Top Twenty Talking Knowledge Leaders:
TOP 20 clearest TV news anchors
The intelligence of Sūrskis and Mauzer’s rookies is yet to be learned
For the sake of interest, Kantar analyzed how artificial intelligence manages to recognize the saying of the less polished knowledge leaders, Sūrskis and Mauzeris of LRT “Dviračio žinios”. Even in the case of these presenters, although they were more wrong, the word recognition rate was around 90%: the robot recognized 92.53%. Speeches by Algis Ramanauskas and 88.21% of characters embodied by Rimas Šapauskas.
“These characters do not shy away from the use of non-normative words, they insert the jargon of others or foreigners, they have a kind of specific pronunciation, so it is natural that such a spicy original language would be more difficult for artificial intelligence to recognize . On the other hand, this is the reason why the system is called intelligence, which learns: after processing a certain amount of dialogues with the mouse, it will also learn to correctly recognize your conversation, ”said an expert from Kantar.
Mauzeris and Sūrskis with Dalia Grybauskaitė / Photo by Robertas Dačkus
Voice recognition will shake the industry
According to the experts of the Kantar Group, voice recognition technologies will have an impact on the market similar to that of smartphones, another great revolution in the industry.
“It is estimated that more than a quarter of searches through Google are done by voice and that it will not be long before that number reaches half, so all companies should be prepared for that. Search By voice will be the new mainstream search and voice technology will become the new today. We started preparing for this a few years ago with the introduction of automatic speech recognition and text-to-text technology for monitoring radio and television news, and now we will start using it to monitor social media transmissions. In the long term, most companies will not be able to do without this technology, “says D. Butkus.
Technology converts speech into text
The innovative knowledge monitoring solution was developed by market research company Kantar in conjunction with language technology company Tilde IT. In addition to this area, this artificial intelligence technology can also be used to translate virtual assistants (“chatbots”) and speak from one language to another (“speech-to-speech”).
Automatic speech recognition technology is based on a deep neural network model that has taught artificial intelligence to recognize words and certain structures in them. More than 250 hours of audio, around 70 million phrases in Lithuanian and more than 700 thousand words were used to train the robots. Particular attention is paid to keywords related to business communication. Neural networks are constantly evolving.
About the study
In the course of this study, Kantar analyzed the speeches of the news anchors of 5 national television networks. Analyzed for 5 min. Knowledge fragments with information read by each presenter. Later, the percentage of WER (word error rate) was calculated: how many words were artificially recognized by artificial intelligence and how many were wrong. The survey was conducted in February this year.
[ad_2]
