TUTORIAL: Speaker Identification with Nemo

A comprehensible and quick illustration on how to do speaker identification on your own data with NVIDIA's Nemo

1. record data
2. data prep
3. config
4. fine tune
5. inference
6. evaluation

Record your own data of speakers intended to be identified

• preprocess into intended format
• slice into around 4 seconds audio samples

Recommended tool pydub

Data Prep (generate train, dev, test manifests)

• instructions

Configration

• change from default config
• documentations & instructions

Fine-tune SpeakerNet on data of speakers intended to be identified

• pre-trained model

• fine tune script

Inference

• inference script

(Evaluation)

• eva