During the pandemic outbreaks, imposed social isolation and quarantine measures cause detrimental psychological mental health problems. Enhancing the Humanoid Robots (HRs) with sophisticated and advanced emotion analysis, speech recognition, and social interaction capabilities can help to ease such problems. This paper proposes a 3-stage speech-based emotion analysis algorithm and then presents its application to a social companion humanoid robot (HR). At the initial stage of the proposed algorithm, most informative speech features are extracted with the Mel-Frequency Cepstral Coefficients (MFCC) technique. Then, in the next stage, each extracted speech features are utilized to determine related emotional condition of the individuals by the Long Short-Term Memory (LSTM) based Recurrent Neural Network (RNN) structure. Finally, in the last stage, human-robot interaction (HRI) is performed with the pre-constructed conversion structures. Although this algorithm can be applied to all HRs, it has been specifically applied to NAO HR in this research. The reliability of the developed algorithm is verified extensively by analyzing each terminal decision about the whole stages with the evaluation of the obtained emotional states from the speech including anger, happiness, surprise, fear, disgust, sadness, and neutrality respectively. Besides all these, the computational burden and accuracy of the algorithm are assessed thoroughly to show the effectiveness of the proposed algorithm.
Anahtar Kelimeler: Human-robot interaction, Long short-term memory, Mel-frequency cepstral coefficients, NAO humanoid robot, Recurrent neural network, Speech-based emotion analysis