Skip to content

README.md

Latest commit

 

History

History
42 lines (32 loc) · 1.31 KB

README.md

File metadata and controls

42 lines (32 loc) · 1.31 KB
Raw

multilingual-phonetic-sv

Language analysis of the VoxCeleb2 dev dataset

We used a language identification model (https://huggingface.co/TalTechNLP/voxlingua107-epaca-tdnn) to analyze the language of the utterances in the VoxCeleb2 dev dataset. The number of utterances for each language can be found in lang.txt.

Comparison of methods

1.Results in this paper

2.Results in Pretraining Conformer with ASR for Speaker Verification

Installation

1.Create a Python 3.9 environment

conda create -n sre python=3.9

2.Pytorch

conda install pytorch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit=11.3 -c pytorch

3.Install required packages

git clone https://github.com/zds-potato/multilingual-phonetic-sv.git
cd multilingual-phonetic-sv
pip -r requirements.txt

Acknowledge

We borrowed a lot of code from:

  1. mfa_conformer
  2. sunine
  3. wespeaker
  4. wenet
  5. ECAPA-TDNN

Other

We will open-source our trained multilingual speech recognition model soon...