Skip to content

VikasOjha666/Speaker-Identification-One-Shot-Learning

Repository files navigation

Speaker-Identification-One-Shot-Learning

This ipython notebook contains the siamese type neural network implementation which calcaulates the similarity scores for a pair of voices. The similarity score is high if the voices belongs to same speaker while score is less for that pair which has voices belonging to different speakers.

The model has been trained on Librispeech dataset and it first converts the audio to spectrogram images and then it compares them using convolutional neural network. This notebook has been created in Google Colab and hence contains linux commands and cannot be executed in anaconda ipython notebook normally.

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published