[3Xin2-13] Quantitative evaluation of utterances in live-streamed content
Keywords:Audio data, Live streaming, Social interaction, Voice recognition, Social media
Live-streaming platforms are a new form of Consumer Generated Media (CGM) and have received much business and academic attention in recent years. Previous studies on live-streaming exist that examined the reasons why audiences watch live-streaming and how streamers should communicate with them. However, these are mainly limited to video game live-streaming platforms. Furthermore, because they use survey data, they do not focus on individual communications between streamers and audiences.
In this study, we investigated a method to quantitatively evaluate each utterance of the streamer using streaming data from the "Lounge," a live voice delivery environment provided by the Japanese music streaming service "AWA". Specifically, we used inaSpeachSegmenter to segment utterances and OpenSmile to extract speech features of the utterances. Then, a machine learning model was built using lightGBM to predict the chat responses of the audiences elicited by the utterances.
This research contributes to the understanding of communication between a live streamer and audiences. In addition, from a practical perspective, the obtained features of live-streaming content can be applied to recommendation use and other applications.
In this study, we investigated a method to quantitatively evaluate each utterance of the streamer using streaming data from the "Lounge," a live voice delivery environment provided by the Japanese music streaming service "AWA". Specifically, we used inaSpeachSegmenter to segment utterances and OpenSmile to extract speech features of the utterances. Then, a machine learning model was built using lightGBM to predict the chat responses of the audiences elicited by the utterances.
This research contributes to the understanding of communication between a live streamer and audiences. In addition, from a practical perspective, the obtained features of live-streaming content can be applied to recommendation use and other applications.
Authentication for paper PDF access
A password is required to view paper PDFs. If you are a registered participant, please log on the site from Participant Log In.
You could view the PDF with entering the PDF viewing password bellow.