# SpeakerRecognition

**Repository Path**: albertyg/SpeakerRecognition

## Basic Information

- **Project Name**: SpeakerRecognition
- **Description**: Deep neural net to identify speakers from a sample of their speech: LibriSpeech corpus.
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2020-06-12
- **Last Updated**: 2020-12-19

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Speaker Recognition

Data comes from the LibriSpeech ASR corpus

http://www.openslr.org/12/

File: train-clean-100.tar.gz (6.3G)(training set of 100 hours "clean" speech) 

5 male speakers and 5 female speakers.

Deep Feed Forward Neural Network built with PyBrain: 

Mean accuracy (over 10 speakers) is 92% with 1 second of voice data, compared to 10% for random guessing. With 3 seconds of voice data, the mean accuracy is 99%. With 200 milliseconds of data, the accuracy is 74%.