Mel-Spectrogram-based ECAPA-TDNN embeddings on Voxceleb

Note: This is a work in progress. This model is intended to be used with a Zero-Shot Multi-Speaker Text-to-Speech (TTS) model available here.

This repository provides all the necessary tools to extract speaker embeddings using a pretrained ECAPA-TDNN model using SpeechBrain. Please note that this model is created for zero-shot multi-speaker TTS, and uses a mel-spectrogram as the input feature. It is trained on Voxceleb 1+ Voxceleb2 training data.

Pipeline description

This system is composed of an ECAPA-TDNN model. It is a combination of convolutional and residual blocks. The embeddings are extracted using attentive statistical pooling. The system is trained with Additive Margin Softmax Loss.

Install SpeechBrain

First of all, please install SpeechBrain with the following command:

pip install speechbrain

Please notice that we encourage you to read our tutorials and learn more about SpeechBrain.

Compute your speaker embeddings

import torchaudio
from speechbrain.inference.encoders import MelSpectrogramEncoder
from speechbrain.utils.fetching import fetch
from speechbrain.utils.data_utils import split_path

spk_emb_encoder = MelSpectrogramEncoder.from_hparams(source="speechbrain/spkrec-ecapa-voxceleb-mel-spec")

INPUT_SPEECH = "speechbrain/asr-wav2vec2-commonvoice-en/example.wav"
source, fl = split_path(INPUT_SPEECH)
path = fetch(fl, source=source, savedir="tmpdir")
ref_signal, fs_file = torchaudio.load(path)
spk_embedding = spk_emb_encoder.encode_waveform(ref_signal)

The system is trained with recordings sampled at 16kHz (single channel).

Inference on GPU

To perform inference on the GPU, add run_opts={"device":"cuda"} when calling the from_hparams method.

Training

The model was trained with SpeechBrain (aa018540). To train it from scratch follows these steps:

Clone SpeechBrain:

git clone https://github.com/speechbrain/speechbrain/

Install it:

cd speechbrain
pip install -r requirements.txt
pip install -e .

Run Training:

cd  recipes/VoxCeleb/SpeakerRec
python train_speaker_embeddings.py hparams/train_ecapa_tdnn_mel_spec.yaml --data_folder=your_data_folder --sample_rate=16000

The training logs will be available here in the future.

Limitations

The SpeechBrain team does not provide any warranty on the performance achieved by this model when used on other datasets.

Referencing ECAPA-TDNN

@inproceedings{DBLP:conf/interspeech/DesplanquesTD20,
  author    = {Brecht Desplanques and
               Jenthe Thienpondt and
               Kris Demuynck},
  editor    = {Helen Meng and
               Bo Xu and
               Thomas Fang Zheng},
  title     = {{ECAPA-TDNN:} Emphasized Channel Attention, Propagation and Aggregation
               in {TDNN} Based Speaker Verification},
  booktitle = {Interspeech 2020},
  pages     = {3830--3834},
  publisher = {{ISCA}},
  year      = {2020},
}

Citing SpeechBrain

Please, cite SpeechBrain if you use it for your research or business.

@misc{speechbrain,
  title={{SpeechBrain}: A General-Purpose Speech Toolkit},
  author={Mirco Ravanelli and Titouan Parcollet and Peter Plantinga and Aku Rouhe and Samuele Cornell and Loren Lugosch and Cem Subakan and Nauman Dawalatabad and Abdelwahab Heba and Jianyuan Zhong and Ju-Chieh Chou and Sung-Lin Yeh and Szu-Wei Fu and Chien-Feng Liao and Elena Rastorgueva and François Grondin and William Aris and Hwidong Na and Yan Gao and Renato De Mori and Yoshua Bengio},
  year={2021},
  eprint={2106.04624},
  archivePrefix={arXiv},
  primaryClass={eess.AS},
  note={arXiv:2106.04624}
}

About SpeechBrain

Website: https://speechbrain.github.io/
Code: https://github.com/speechbrain/speechbrain/
HuggingFace: https://huggingface.co/speechbrain/