README.md

metadata

license: mit
datasets:
  - s3prl/mini_voxceleb1
language:
  - en
metrics:
  - accuracy
pipeline_tag: audio-classification
tags:
  - speech
  - text
  - cross-modal
  - unified model
  - self-supervised learning
  - SpeechT5
  - Speaker Identification
  - Speaker Recognition

SpeechT5 SID

| Github | Huggingface |

This manifest is an attempt to recreate the Speaker Identification recipe used for training SpeechT5. This manifest was constructed using VoxCeleb1 containing over 100,000 utterances for 1,251 celebrities. The identification split are given as follows.

	train	valid	test
# of speakers	1,251	1,251	1,251
# of utterances	138,361	6,904	8,251

Requirements

• Fairseq

Tools

• manifest/utils is used to produce manifest as well as conduct training, validation, and evaluation.
• mainfest/iden_split.txt and mainfest/vox1_meta.csv are officially released files.

Model and Results

• speecht5_sid.pt are reimplemented Speaker Identification fine-tuning on the released manifest but with a smaller batch size (Ensure the manifest is ok).
• results are reproduced by the released fine-tuned model and the accuracy is $96.194%$.
• log is the tensorboard log of fine-tuning the released model.

Reference

If you find our work is useful in your research, please cite the following paper:

@inproceedings{ao-etal-2022-speecht5,
    title = {{S}peech{T}5: Unified-Modal Encoder-Decoder Pre-Training for Spoken Language Processing},
    author = {Ao, Junyi and Wang, Rui and Zhou, Long and Wang, Chengyi and Ren, Shuo and Wu, Yu and Liu, Shujie and Ko, Tom and Li, Qing and Zhang, Yu and Wei, Zhihua and Qian, Yao and Li, Jinyu and Wei, Furu},
    booktitle = {Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
    month = {May},
    year = {2022},
    pages={5723--5738},
}