Part 1: Automated Speech-Emotions Analysis Tool for Autism Children
Hello everyone, Today we are going to develop a speech recognition system to identify the impairments of children. The main idea behind this work is to identify the child’s emotions when the child is talking.
This article is part of a series.
The focus of this study was to develop an emotion tracking system for classifying unusual emotions of autistic children while the child is talking.
Also by identifying these abnormal behaviors in emotions and providing some therapies as the treatment of autism.
Moreover, it will help speech pathologists to verify their second thoughts from the system-generated report.
Let’s look at the main functions of this developed web page.
Upload Records — In this work, I applied the manual file-uploading method to analyze emotion. After the manually uploaded recordings, it automatically analyses the audio without any instruction. At the end of the operation, it displayed the output in a string format. Real-time
Real-time check — For an excellent basis for solving the task, I implemented a method to test voice in real time. After the “Please talks” prompt has appeared, the child’s speech will be recorded, and when the child stops speaking recording is automatically stopped. And it generates a report.
Exercise / Patient Test — Here we implemented several therapies (Execrices) to treat children.
SUMMARY:
There are special kinds of children that have narrow interests, communication difficulties, and repetitive behaviors. They were called autistic children and symptoms of this state can be noticed in early childhood.
One of the main diagnostic characteristics of Autism spectrum disorder is the child's unusual emotions and expressions.
From this project, emotions are classified automatically capturing speech while a child is talking. Additionally, the achievements were classifying emotion with language-independent, noise reduction, and age-independence features.
The input audio stream of children was normalized into a specific range, sub-framed into 2s length, and extracted the most effective 40 audio features.
Then feeding to the deep neural network (CNN) and the trained model has the ability to classify eight different emotions:
Sad, Surprise, Neutral, Happy, Calm, Fear, Disgust, and Angry even in an uncontrol environment.
Even if the classified emotions have small frequency variances, the trained CNN model has the ability to handle them.
The model has achieved an F1 score of 0.90.
Moreover, this study aims to verify the second thoughts of identifying the autism diagnosis of children.
This article is part of a series and Here is my plan:
Part 1: What we are going to do — Summary
Part 2: What is Autism spectrum disorder (ASD) and the role of identifying emotion in speech impairments?
Part 3: System Overview / Data source and Dataset preparation.
Part 4: Software Solution / Feature Extraction & Selection (MFCC, Chromagram, MEL, etc..)
Part 5: Emotion Detection Algorithm Selection.
Part 6: Final Emotion Recognition (Deep Learning)
Part 7: Testing and Implementation.
Part 8: Results, Findings, and Discussion.
Part 9: Conclusion.
My next article is regarding: What is Autism spectrum disorder (ASD) and the role of identifying emotion in speech impairments?
If you enjoyed this piece, please hit the clap button 👏
See you guys in the next article! ❤️ ✌