GenAID

This is the repository for GenAID, a Generalisable accent identification (AID) model across speakers. The code is built upon speechbrain v0.5.16 with the original documentation here, and the AID dataset construction and code implementation by CommonAccent.

I. Environment Setup

git clone https://github.com/jzmzhong/GenAID.git
cd GenAID
conda create -n speechbrain python==3.10
conda activate speechbrain
conda install pytorch==2.3.0 torchvision==0.18.0 torchaudio==2.3.0 pytorch-cuda=11.8 -c pytorch -c nvidia
pip install -r requirements.txt
pip install --editable .

II. Data Preparation

Download Common Voice

1. Download the dataset, which is available here. The specific data version used by this repo is Common Voice Corpus 17.0.

2. Unzip the dataset after download.

   tar -xvf cv-corpus-17.0-2024-03-15-en.tar.gz

Process into Common Accent

1. Filter out the speech utterances with accent labels and process them.

2. Filter out accents with sufficient data and split into training/validation/testing sets. Note that there are two validation/testing sets, one for seen speakers (these speakers have sufficient data in the training set) and the other for unseen speakers (these speakers do not overlap with any of the speakers in the training set).

3. Processed training/validation/testing sets are available at:

   ./recipes/CommonAccent/CommonAccent-CV17-spk-resplit.

III. Training

Modify the Paths in the Configuration File

1. Please ensure that the data_folder field is the Common Voice dataset directory, and the csv_prepared_folder field is the Common Accent processed training/validation/testing sets directory, e.g. ./recipes/CommonAccent/CommonAccent-CV17-spk-resplit.

2. Also set the output_folder field to be the directory where you want to store the trained checkpoints and the rir_folder filed to be the directory where you want to store the noise dataset, downloaded and used in training for more robuts accent identification.

Run the Model Training Script

cd ./recipes/CommonAccent
python train_GenAID.py train_GenAID_v6.yaml

Trained Model

1. A trained model (with partial files for inference) is available at: https://drive.google.com/file/d/1slGrpZSu5g-nF7R-QMCmtGcjN3kw7lQj/view?usp=sharing

2. Please download and unzip it into the following directory for inference/embeddings extraction.

./recipes/CommonAccent/GenAID_v7

IV. Inference

Modify the Paths in the Configuration File

1. Please set pretrained_path field to be the checkpoint directory you want to inference, and the output_folder field to be the directory where you want to store the inference results (confusion matrices).

Run the Model Inference Script

cd ./recipes/CommonAccent
python inference_GenAID.py inference_GenAID_v6.yaml

Inference Results

Inference results are available here:

./recipes/CommonAccent/results

V. Embeddings Extraction

Data Preparation

1. Please process the data to extract embeddings from and place the original data under data_folder, and the csv file of all files information under csv_prepared_folder. An example of the data processing scripts is at ./recipes/CommonAccent/VCTK/data_prep_VCTK.py, with processed results at ./VCTK/all_file_paths.csv.

Modify the Paths in the Configuration File

1. Please process the data to extract embeddings from and place the original data under data_folder, and the csv file of all files information under csv_prepared_folder. An example of the data processing scripts is at ./recipes/CommonAccent/VCTK/data_prep_VCTK.py, with processed results at ./VCTK/all_file_paths.csv.

2. Please set pretrained_path field to be the checkpoint directory you want to use to extract embeddings, and the output_folder field to be the directory where you want to store the extracted embeddings.

Run the Embedding Extraction Script

cd ./recipes/CommonAccent
python extract_embeddings_GenAID.py extract_embeddings_GenAID_v6.yaml

VI. Reference

CommonAccent: Paper, Code, Model

VI. Citing

Please cite GenAID (part of the AccentBox paper) if you use it for your research or business.

@inproceedings{zhong2025accentbox,
    author = {Zhong, Jinzuomu and Richmond, Korin and Su, Zhiba and Sun, Siqi},
    title = {{AccentBox: Towards High-Fidelity Zero-Shot Accent Generation}},
    booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
    year = {2025},
    pages = {1-5},
    doi = {10.1109/ICASSP49660.2025.10888332}
}