Zero-sample voice emotion recognition method based on prototype reconstruction and generative learning

• Main Authors: DENG JUN, ZHANG ZIXING, YIN ZHIPENG, CHEN YONGFA, XU XINZHOU, YANG ZHEN
• Format: Patent
• Language: Chinese; English
• Published: 14.07.2023
• Subjects: ACOUSTICS; CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; MUSICAL INSTRUMENTS; PHYSICS; SPEECH ANALYSIS OR SYNTHESIS; SPEECH OR AUDIO CODING OR DECODING; SPEECH OR VOICE PROCESSING; SPEECH RECOGNITION

The invention belongs to the field of voice signal emotion recognition, and discloses a zero-sample voice emotion recognition method based on prototype reconstruction and generation learning. The method comprises the following steps: firstly, extracting sublingual features from a known emotion category phrase sample, training in combination with a known emotion category semantic embedding prototype and a known emotion category phrase sample label to obtain an optimal prototype reconstruction model, and respectively obtaining a known emotion category reconstruction prototype and an unknown emotion category reconstruction prototype in combination with a known emotion category semantic embedding prototype and an unknown emotion category semantic embedding prototype; then, extracting paragraph features from the known emotion category phrase samples, according to known emotion category phrase sample labels, combining a known emotion category reconstruction prototype and training to obtain an optimal generation lea