|
--- |
|
language: "en" |
|
thumbnail: |
|
tags: |
|
- speechbrain |
|
- embeddings |
|
- Speaker |
|
- Verification |
|
- Identification |
|
- pytorch |
|
- ECAPA-TDNN |
|
license: "apache-2.0" |
|
datasets: |
|
- voxceleb |
|
metrics: |
|
- EER |
|
- Accuracy |
|
widget: |
|
- example_title: VoxCeleb Speaker id10003 |
|
src: https://cdn-media.huggingface.co/speech_samples/VoxCeleb1_00003.wav |
|
- example_title: VoxCeleb Speaker id10004 |
|
src: https://cdn-media.huggingface.co/speech_samples/VoxCeleb_00004.wav |
|
--- |
|
|
|
# Speaker Identification with ECAPA-TDNN embeddings on Voxceleb |
|
|
|
This repository provides a pretrained ECAPA-TDNN model using SpeechBrain. The system can be used to extract speaker embeddings as well. Since we can't find any resource that has SpeechBrain or HuggingFace compatible checkpoints that has only been trained on VoxCeleb2 development data, so we decide to pre-train an ECAPA-TDNN system from scratch. |
|
|
|
# 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. |
|
|
|
We use FBank (16kHz, 25ms frame length, 10ms hop length, 80 filter-bank channels) as the input features. It was trained using initial learning rate of 0.001 and batch size of 512 with cyclical learning rate policy (CLR) for 10 epochs on 4 A100 GPUs. We employ additive noises and reverberation from [MUSAN](http://www.openslr.org/17/) and [RIR](http://www.openslr.org/28/) datasets to enrich the supervised information. The pre-training progress takes approximately seven days for the ECAPA-TDNN model. |
|
|
|
# Performance |
|
|
|
| Splits | Backend | S-norm | EER(%) | minDCF(0.01) | |
|
|:-------------:|:--------------:|:--------------:|:--------------:|:--------------:| |
|
| VoxCeleb1-O | cosine | no | 1.45 | 0.17 | |
|
| VoxCeleb1-E | cosine | no | TBD | TBD | |
|
| VoxCeleb1-H | cosine | no | TBD | TBD | |
|
|
|
# Compute the speaker embeddings |
|
|
|
The system is trained with recordings sampled at 16kHz (single channel). |
|
|
|
```python |
|
import torchaudio |
|
from speechbrain.pretrained import EncoderClassifier |
|
classifier = EncoderClassifier.from_hparams( |
|
source="yangwang825/ecapa-tdnn-vox2" |
|
) |
|
signal, fs = torchaudio.load('spk1_snt1.wav') |
|
embeddings = classifier.encode_batch(signal) |
|
``` |
|
|
|
You can find our training results (models, logs, etc) here. |
