Abstract
In this article, we study emotion detection from speech in a speaker-specific scenario. By parameterizing the excitation component of voiced speech, the study explores deviations between emotional speech (e.g., speech produced in anger, happiness, sadness, etc.) and neutral speech (i.e., non-emotional) to develop an automatic emotion detection system. The excitation features used in this study are the instantaneous fundamental frequency, the strength of excitation and the energy of excitation. The Kullback-Leibler (KL) distance is computed to measure the similarity between feature distributions of emotional and neutral speech. Based on the KL distance value between a test utterance and an utterance produced in a neutral state by the same speaker, a detection decision is made by the system. In the training of the proposed system, only three neutral utterances produced by the speaker were used, unlike in most existing emotion recognition and detection systems that call for large amounts of training data (both emotional and neutral) by several speakers. In addition, the proposed system is independent of language or lexical content. The system is evaluated using two databases of emotional speech. The performance of the proposed detection method is shown to be better than that of reference methods.
Highlights
In addition to its linguistic contents, speech contains rich information about the speaker, such as the gender, age and emotional state
SUMMARY AND CONCLUSION In this study, we proposed an automatic emotion detection system from speech using excitation features extracted around glottal closure instants (GCIs)
Using the KL distance, the system measures the deviation between the reference utterance produced in a neutral state and a test utterance of emotional speech
Summary
In addition to its linguistic contents, speech contains rich information about the speaker, such as the gender, age and emotional state. Alku: Excitation Features of Speech for Speaker-Specific Emotion Detection (e.g., a convolutional neural network (CNN) or a bidirectional long short-term memory (BLSTM) network) is trained to conduct the recognition task directly from the input (either from the raw signal waveform or from the spectrogram) [20], [21]. Both of these two approaches, are data driven and they call for lots of training data [16]–[21].
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Disclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.