search

shaharpit809 / Audio-Sentiment-Analysis

This repository consists of work done to analyse sentiment of a customer in a conversation with a call center agent using various machine learning algorithms and audio features.
79 stars 40 forks source link

dataset of call recording and ravdess is different #1

Closed dimanshu closed 4 years ago

dimanshu commented 4 years ago

dataset of call recording and ravdess is different it will create the problem ? and yes will it solve the problem to not catch high pitch sound as anger. like generally i speak very loudly so it will detect my voice as anger ?

shaharpit809 commented 4 years ago

Hi @dimanshu ,

Thank you for your message. Having different datasets from different dialects does create problems. It is preferable to use data that is similar to the training data to capture the emotion in the most efficient manner. Again, the model will output how you train it. If the training data has loud emotion for angry audio clips then it will consider loud sound as anger. You can try using more audio features and check if its working or not.