Automatic Speech Recognition for Javanese Language

Introduction

Automatic Speech Recognition is a model deep learning to convert speech into text. This model is trained using the OpenSLR dataset. The dataset used is the Javanese language dataset. The dataset is divided into 2 parts, namely train, and validation. The dataset is trained using the wav2vec2 based model from Facebook.

How to use

1. Clone this repository

git clone https://github.com/JohanesSetiawan/asr-javanese-api.git

2. Install requirements

pip install Flask flask-cors huggingsound pyngrok

if you using server, and not the local machine.

or

pip install Flask flask-cors huggingsound

if you using local machine.

3. Run the server

python api.py

4. Paste folder webui to XAMPP or Laragon

Paste the API link that appears in the terminal to the transcribe.php file, in the:

url: "<FLASK_API>/transcribe", // Flask server API endpoint variable.

and the endpoint will be:

/transcribe

6. Open the webui

7. Upload the audio file

8. Click the Transcribe button

9. Wait for the process to finish

10. The result will appear in the Transcription section

References

Javanese Model

Base Model

Training Code

You can use API in locally or using Google Colab to run API.