Speaker Recognition & Verification Experiments with Common Voice.

This repository contains modified scripts of SpeechBrain for running speaker identification and verification experiments with the Persian Common Voice dataset. State-of-the-art ECAPA-TDNN model is utilized for this experiment.

The original scripts for training on VoxCeleb are here.

Speaker verification using ECAPA-TDNN embeddings

The following notebooks used to train, test, and use ECAPA-TDNN model:

• Train notebook
• Test notebook
• Demo notebook

Not all file paths used in the notebooks are available publicly.

Performance summary

[Speaker Verification Results with Common Voice]

System	Dataset	Accuracy1	Stress Test Accuracy2	Model Link3
ECAPA-TDNN	Persian Common Voice	97.5%	86.8%	https://drive.google.com/drive/folders/1R_gvC_St56Atxfu8MLRb1PIlBnBahta2?usp=sharing

¹ Tested on a private hand-made dataset, consisting of 8400 pair samples.

² Tested on German BROTHERS dataset.

³ Model is not temporarily available for public use.

Pretrained Model + Easy-Inference

You can find the pre-trained model on Google Drive. Moreover, an easy-inference interface is available here.