spkrec ecapa voxceleb

Introduction

This repository provides tools for speaker verification using a pretrained ECAPA-TDNN model from SpeechBrain. The model is trained on the VoxCeleb dataset and is capable of extracting speaker embeddings and performing verification tasks. It achieves an Equal Error Rate (EER) of 0.80% on the VoxCeleb1-test set.

Architecture

The system utilizes an ECAPA-TDNN model, which combines convolutional and residual blocks. Embeddings are extracted through attentive statistical pooling, and the model is trained using Additive Margin Softmax Loss. Speaker verification is accomplished by computing the cosine distance between speaker embeddings.

Training

The model was trained using SpeechBrain with recordings sampled at 16kHz. To train from scratch:

1. Clone the SpeechBrain repository.

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

2. Install the required packages.

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

3. Execute the training script.

   cd recipes/VoxCeleb/SpeakerRec
   python train_speaker_embeddings.py hparams/train_ecapa_tdnn.yaml --data_folder=your_data_folder
   

Guide: Running Locally

1. Install SpeechBrain:

   pip install git+https://github.com/speechbrain/speechbrain.git@develop
   

2. Compute Speaker Embeddings:

   import torchaudio
   from speechbrain.inference.speaker import EncoderClassifier
   classifier = EncoderClassifier.from_hparams(source="speechbrain/spkrec-ecapa-voxceleb")
   signal, fs = torchaudio.load('tests/samples/ASR/spk1_snt1.wav')
   embeddings = classifier.encode_batch(signal)
   

3. Perform Speaker Verification:

   from speechbrain.inference.speaker import SpeakerRecognition
   verification = SpeakerRecognition.from_hparams(source="speechbrain/spkrec-ecapa-voxceleb", savedir="pretrained_models/spkrec-ecapa-voxceleb")
   score, prediction = verification.verify_files("tests/samples/ASR/spk1_snt1.wav", "tests/samples/ASR/spk2_snt1.wav")
   

4. Inference on GPU: Add run_opts={"device":"cuda"} when calling from_hparams.

Cloud GPUs

For optimal performance, consider using cloud GPU services such as AWS, Google Cloud, or Azure for running computations.

License

This project is licensed under the Apache-2.0 License.