Accent Classification Brazillian Portuguese using CNN models

This repository showcases a classification task for Brazilian Portuguese using two model configurations: a 1D Convolutional Neural Network (CNN) combined with Long Short-Term Memory (LSTM), and a standalone 2D CNN. The CNN1D + LSTM model, based on Tostes' work, utilizes a range of frequency values from a spectrogram as input. Meanwhile, the CNN2D model processes images sized at 227x227 pixels.

Models

Convolution 1D with LSTM

Convolution 2D

Datasets

Two subsets of Spotify Podcasts ( Spotify A and Spotify B) more informations is presented in following repository: Spotify Subsets

Experiments

• Local :

git clone https://github.com/aryamtos/accent-classification-audio.git
pip3 install -r requirements.txt

• Conda Environment 🐍

git clone https://github.com/aryamtos/accent-classification-audio.git
conda create --name myenv
conda install --file requirements.txt
conda list

Build Docker image 🐳

docker build -t accent:2.0 .
docker images

Run Container Docker

docker run -it --gpus all --ulimit memlock=-1 --ulimit stack=67108864 -v vol/:/vol/ --name accentBr -d accent:2.0
docker exec -it accentBr /bin/bash