机器人>> speech-accent-classifier>> 返回
项目作者： stephenjkaplan

项目描述 ：
Classifying audio of human speech into various accents/countries of origin.
高级语言： Jupyter Notebook
项目地址: git://github.com/stephenjkaplan/speech-accent-classifier.git
创建时间： 2020-07-25T03:43:38Z
项目社区：https://github.com/stephenjkaplan/speech-accent-classifier
开源协议：
下载


Speech Accent Classifier

Play around with the model and make predictions using the web app
I made using Flask.

Read the blog post.

Description

Classifying audio of human speech into various accents/countries of origin using
MFCC coefficients extracted from audio .wav files.

The model in flask_app/static/sklearn_models/final_model.pkl is an ensemble of K-Nearest Neighbor and Logistic
Regression models. The overall predictive accuracy of the model is 0.89 and it has an ROC AUC score of 0.95. The
blog post about it is here.

This was developed over a 2-week span in August 2020 as a project for the Metis data science
program.

Data Source

Fokoue, E. (2020). UCI Machine Learning Repository - Speaker Accent Recognition Data Set. Irvine, CA: University of California, School of Information and Computer Science.

File Contents

  • Running main.py will launch the Flask app locally.
  • utilities.py contains files needed for the Flask app to run.
  • notebooks/ contains the Jupyter Notebook used to do all data analysis and modeling, as well as an accompanying
    file of Python functions.
  • templates/ contains HTML files for the Flask application
  • static/ contains static files for the Flask application as well as the pickled scikit-learn model.
  • heroku.yml and Dockerfile are used for deployment of the Flask app to Heroku

Dependencies

The contents of /notebooks can’t be fully run because the Jupyter Notebook connects to a remote SQL database. In order
to install the dependencies for the flask app, run:

pip install -r requirements.txt