4. Method
1. Extract and label known instances of names from dataset
2. Build CNN-based name recognition model
3. Identify phonetically similar instances from transcription
4. Run model on the identified similar instances
5. Training a Convolutional Neural Network (CNN)
Trained
Model
Learns Features
Converts to Spectrogram
Labeled Data (E.g. Biden)
7. Next Steps
1. Build CNN-based name recognition model to work on Joe Biden
and Donald Trump and test accuracy of the model
2. Build new instances of data by adding background noise to
pre-existing name instances and use Google Text To Speech to
create more instances.
3. Gather audio data for uncommon candidate names and retrain the
CNN-based name recognition model.