Psychological State Detection Using AI
In our increasingly digital world, effective communication with machines has become integral to our daily lives. However, a significant challenge lies in bridging the emotional gap between humans and artificial intelligence. Traditional human-computer interfaces often miss the nuanced emotional cues present in our voices, hindering our ability to interact with machines in a more natural and emotionally intelligent way.
By using Tensorflow, Streamlit, and LSTM
We trained our model on a huge audio datasets of 200 target words were spoken in the carrier phrase "Say the word _' by two actresses and recordings were made of the set portraying each of seven emotions (anger, disgust, fear, happiness, pleasant surprise, sadness, and neutral) with a total of 2800 data points.
Feature Extraction: extracting relevant features from audio data. These features include pitch, tone, intensity, and spectral characteristics.
Model Training: LSTM networks, as part of the TensorFlow framework, are trained on labeled audio datasets that associate audio samples with specific psychological states (e.g., happiness, sadness, anger).
Pattern Recognition: During training, the LSTM learns to recognize patterns in the extracted audio features that correlate with different psychological states. It identifies how changes in vocal attributes correspond to specific emotions.
Inference by Streamlit: Once trained, the AI model can infer the psychological state of unseen audio data. It analyzes the audio's features and provides an estimation of the emotional state expressed in the speech.
