Speech Analyser 🎤

by Aditya Kulshrestha — 17/1/2023

Introduction

This year in Foss Overflow 2022, I worked on Speech Analyser 🎤 project. People who are learning to speak English may make mistakes when they communicate. In order to assist people talk more effectively, we want to provide them feedback on their speech. Speech Analyzer project aims to help people improve their communication skills and become more confident and effective speakers. Our goal is to provide an easy-to-use solution that can analyze speech patterns and provide real-time feedback, enabling users to identify and eliminate disfluencies.

How was my experience with FOSSOverflow'22?📈

Participating in this program greatly increased my confidence as a contributor to the open-source community. It taught me how to approach and solve problems effectively, by breaking them down into key steps. I gained knowledge about various techniques and libraries in the natural language processing domain, and also learned how to build an end-to-end NLP project. I was aware that this project would require a significant amount of effort. Despite facing difficulties along the way, with the help of my co-contributor Ananya Srivastava and my mentors Ananya and Wassim Chouchen I was able to complete it.

This program has assisted me in understanding the open-source development process and community dynamics, gaining experience working on real-world projects with other contributors. Additionally, it has also helped me to build a portfolio of open-source contributions that demonstrate my abilities and improve my communication and collaboration skills, which are essential for working in an open-source community.

Work Done

I accomplished the following tasks during the program.

  1. Searching for multiple Automatic Speech Recognition libraries and tools that detects disfluencies 🔎.
  2. Data Collection and preprocessing for disfluency recognition 📑.
  3. Annotation of the dataset and checking for quality of data 📝.
  4. Fine-tuning a NER Model with 95% accuracy.
  5. Pushing of the model on HuggingFace for easy inference .
  6. Building a gradio app for inferencing of ASR, disfluency detection and grammar check .

Challenges Faced

During the project I faced a lot of challenges. One was to collect dataset according to the project's requirement. The right data with good quality was an important aspect of this project. I also faced difficulties while building a model that would detect disfluencies. But thanks to my mentors, I was able to clear all the difficulties that I faced during the project, with their guidance and support being instrumental in the projects' success.

Participating in FOSS Overflow is an excellent chance for aspiring open-source contributors to gain insight into the workings of the open-source community, sharpen their skills, and build confidence. Overall, it is a valuable opportunity for individuals looking to get involved in open-source development.