1. Emotion Detection Engine for mixed languages a) Emotion detection model trained for a mix of Cantonese with English b) Emotion analytics of 4 different emotions c) Detection accuracy rate is >75% 2. Punctuator for sentence meaning extraction a) Punctuation placement in a sentence: ".", "?", ",", and "!". b) Accuracy rate is >80% 3. Text to Speech Synthesizer a) Human like computer generated voice based on Deep Learning Algorithm b) Can speak Cantonese with some common English words in the same sentence c) Feasibility study on how to express proper emotions in speech 4. AI Chatbot for Telephone Customer Service with Emotion Detection a) Removal of background noise and background music on incoming speech b) Emotion detection on incoming speech c) Speech generation with emotion d) Dialogue System CS Deliverables for Customer 1: 1.1: Speech Recognition Engine (SRE) running on Windows 10 Tablets 1.2: Integration of SRE to Customer’s existing web browser application 1.3: Software on natural language processing (NLP) to extract keywords on inventory enquiries 1.4: Software for end-to-end integration of Customer’s web application and NLP 1.5: Enquiry Management System 1.6: Software for housekeeping of audio files on Windows 10 tablets 1.7: Documentation (installation guide and test plan & test report) CS Deliverables for Customer 2: 2.1. Software for Speech Recognition Engine (SRE) running on a standalone server 2.2. A software graphical user interface (GUI) to upload files for processing 2.3. A software graphical user interface (GUI) to show record and statistics of the processed file 2.4 Retraining of SRE
Bibo Limited (Contract Service) [Sponsor]
Chow Tai Fook Jewellery Company Limited (Contract Service) [Sponsor]
A sentence, when spoken with different intonations, can convey different meanings. The emotion or sentiment lies more in the subtlety of the tone than the text itself. Without extracting the proper emotion but based solely on the text, misunderstanding can arise. A lot of research has been conducted to detect the emotion in speech for major languages such as English or Putonghua. Because of the scarcity of data in Cantonese, emotion detector for Cantonese is not readily available. In this project, an emotion detector to determine emotional states of Cantonese speakers will be developed. With the correct emotion identified, the most appropriate answer can be provided by the chatbot. However, when the answer is given without the injection of emotion, a sense of apathy will be misinterpreted, which could arouse unnecessary emotion in the listeners. Therefore an emotional speech synthesis system for mixed languages (Cantonese with English words) will be studied to replace the monotonous, robotic speech synthesizer.