README.md

---
license: apache-2.0
---
# UK & Ireland Accent Classification Model

This is a model to classify and identify the accent of a UK or Ireland speaker among one of the following accents:
* Irish English
* Midlands English
* Northern English
* Scottish English
* Southern English
* Welsh English

The model implements transfer learning feature extraction using [Yamnet](https://tfhub.dev/google/yamnet/1) model in order to train a model.

## Yamnet Model
Yamnet is an audio event classifier trained on the AudioSet dataset to predict audio events from the AudioSet ontology. It is available on TensorFlow Hub.
Yamnet accepts a 1-D tensor of audio samples with a sample rate of 16 kHz.   
As output, the model returns a 3-tuple:   
- scores of shape (N, 521) representing the scores of the 521 classes
- embeddings of shape (N, 1024)
- log_mel spectrogram representing the log-mel spectrogram of the entire audio frame
We will use the embeddings, which are the features extracted from the audio samples, as the input to our dense model. 

## Dense Model
The dense model that we used consists of:
- An input layer which is embedding output of the Yamnet model
- 4 Dense hidden layers and 4 Dropout layers
- An output dense layer

## Dataset
The dataset used is the **[Open-source Multi-speaker Corpora of the English Accents in the British Isles](https://openslr.org/83/)** which consists of a total of **17,877 audio files**.

### Dataset Info
@inproceedings{demirsahin-etal-2020-open,  
title = {{Open-source Multi-speaker Corpora of the English Accents in the British Isles}},  
author = {Demirsahin, Isin and Kjartansson, Oddur and Gutkin, Alexander and Rivera, Clara},  
booktitle = {Proceedings of The 12th Language Resources and Evaluation Conference (LREC)},  
month = may,  
year = {2020},  
pages = {6532--6541},  
address = {Marseille, France},  
publisher = {European Language Resources Association (ELRA)},  
url = {https://www.aclweb.org/anthology/2020.lrec-1.804},\n\ ISBN = {979-10-95546-34-4},  
}

# Demo
A demo is available in HuggingFace Spaces ...