This model extracts speaker embeddings from given speech, which is the backbone for speaker verification and diarization tasks.
It is a "large" version of TitaNet (around 23M parameters) models.
See the model architecture section and [NeMo documentation](https://docs.nvidia.com/deeplearning/nemo/user
How to Use this Model
The model is available for use in the NeMo toolkit [3] and can be used as a pre-trained checkpoint for inference or for fine-tuning on another dataset.
Automatically instantiate the model
python
1import nemo.collections.asr as nemo_asr
2speaker_model = nemo_asr.models.EncDecSpeakerLabelModel.from_pretrained("nvidia/speakerverification_en_titanet_large")
This model accepts 16000 KHz Mono-channel Audio (wav files) as input.
Output
This model provides speaker embeddings for an audio file.
Model Architecture
TitaNet model is a depth-wise separable conv1D model [1] for Speaker Verification and diarization tasks. You may find more info on the detail of this model here: TitaNet-Model.
Training
The NeMo toolkit [3] was used for training the models for over several hundred epochs. These model are trained with this example script and this base config.
Datasets
All the models in this collection are trained on a composite dataset comprising several thousand hours of English speech:
Voxceleb-1
Voxceleb-2
Fisher
Switchboard
Librispeech
SRE (2004-2010)
Performance
Performances of the these models are reported in terms of Equal Error Rate (EER%) on speaker verification evaluation trial files and as Diarization Error Rate (DER%) on diarization test sessions.
Speaker Verification (EER%)
| Version | Model | Model Size | VoxCeleb1 (Cleaned trial file) |
|---------|--------------|-----|---------------|
| 1.10.0 | TitaNet-Large | 23M | 0.66 |
Speaker Diarization (DER%)
| Version | Model | Model Size | Evaluation Condition | NIST SRE 2000 | AMI (Lapel) | AMI (MixHeadset) | CH109 |
|---------|--------------|-----|----------------------|---------------|-------------|------------------|-------|
| 1.10.0 | TitaNet-Large | 23M | Oracle VAD KNOWN # of Speakers | 6.73 | 2.03 | 1.73 | 1.19 |
| 1.10.0 | TitaNet-Large | 23M | Oracle VAD UNKNOWN # of Speakers | 5.38 | 2.03 | 1.89 | 1.63 |
Limitations
This model is trained on both telephonic and non-telephonic speech from voxceleb datasets, Fisher and switch board. If your domain of data differs from trained data or doesnot show relatively good performance consider finetuning for that speech domain.
NVIDIA Riva: Deployment
NVIDIA Riva, is an accelerated speech AI SDK deployable on-prem, in all clouds, multi-cloud, hybrid, on edge, and embedded.
Additionally, Riva provides:
World-class out-of-the-box accuracy for the most common languages with model checkpoints trained on proprietary data with hundreds of thousands of GPU-compute hours
Best in class accuracy with run-time word boosting (e.g., brand and product names) and customization of acoustic model, language model, and inverse text normalization
Streaming speech recognition, Kubernetes compatible scaling, and enterprise-grade support
License to use this model is covered by the CC-BY-4.0. By downloading the public and release version of the model, you accept the terms and conditions of the CC-BY-4.0 license.
