[4Xin1-53] A study of acoustic features in inside sales speech
Keywords:Inside speech, speech recognition, dimension reduction
(1) Purpose
The success of a business conversation is influenced by the acoustic impression of the speech, in addition to its linguistic context. We are studying the acoustic characteristics of inside sales speech in order to estimate sales performance and to train the vocalizations for business conversations.
Related research has proposed acoustic features that can be used for emotion estimation (e.g., eGMAPS). On the other hand, studies on business conversation are mainly interested in linguistic features of speech, and only a few acoustic features of speech used in business conversations have been studied.The purpose of this study is to select important acoustic features in business conversations.
(2) Results
The distribution of acoustic features was examined by acquiring simulated business meeting speech with controlled linguistic context. The 88 dimensional acoustic features proposed by Eyben et al. were calculated comprehensively. In order to find the acoustic features that reveal speaker differences, these acoustic features were visualized using the t-SNE method and the K-means method.
The results showed that the mean of the delta F0, the variance of the delta power, and the mean of the strength of vocal fold vibration characterized the voice of the inside sales.
The success of a business conversation is influenced by the acoustic impression of the speech, in addition to its linguistic context. We are studying the acoustic characteristics of inside sales speech in order to estimate sales performance and to train the vocalizations for business conversations.
Related research has proposed acoustic features that can be used for emotion estimation (e.g., eGMAPS). On the other hand, studies on business conversation are mainly interested in linguistic features of speech, and only a few acoustic features of speech used in business conversations have been studied.The purpose of this study is to select important acoustic features in business conversations.
(2) Results
The distribution of acoustic features was examined by acquiring simulated business meeting speech with controlled linguistic context. The 88 dimensional acoustic features proposed by Eyben et al. were calculated comprehensively. In order to find the acoustic features that reveal speaker differences, these acoustic features were visualized using the t-SNE method and the K-means method.
The results showed that the mean of the delta F0, the variance of the delta power, and the mean of the strength of vocal fold vibration characterized the voice of the inside sales.
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.