Zwart, Tessa (2023) Sarcastic speech synthesis in Dutch using voice-transformation. Master thesis, Voice Technology (VT).
|
PDF
MA 3683850 T Zwart.pdf Download (431kB) | Preview |
Abstract
This research delves into the creation of a voice that sounds sarcastic and examines the recognition of sarcasm in synthetic speech. Sarcasm, known for its ability to convey meaning beyond literal interpretation, plays a significant role in our everyday interactions. Our main focus lies in identifying the acoustic features relevant to Dutch sarcasm, utilizing the FastSpeech2 model for synthesizing and manipulating speech. To evaluate the synthesized speech, a survey was conducted, which revealed that the sarcastic synthetic voice has a limited recognition accuracy of up to 35%. This suggests that the factors responsible for conveying sarcasm are not collectively perceived as sarcastic by listeners. Nevertheless, when the manipulation of the speech files is doubled, the accuracy rises to as high as 43%, suggesting that increasing the degree of manipulation leads to a higher likelihood of recognizing the speech as sarcastic. Additionally, we investigated three sentence types (tag-questions, declaratives, and wh-exclamatives) and found no significant difference in recognizing sarcasm based on sentence type. This implies that sentence structure does not influence people's perception of sarcasm. Potential explanations for the low recognition rates include the need for further modifications to synthetic speech or the incorporation of facial expressions to enhance sarcasm recognition.
| Item Type: | Thesis (Master) |
|---|---|
| Name supervisor: | Coler, M.L. and Gao, X. |
| Date Deposited: | 12 Sep 2023 11:14 |
| Last Modified: | 12 Sep 2023 11:14 |
| URI: | https://campus-fryslan.studenttheses.ub.rug.nl/id/eprint/374 |
Actions (login required)
 |
View Item |