INDONESIAN AUTOMATIC SPEECH RECOGNITION MODEL TRAINING USING LOW RANK ADAPTATION METHOD
This research focuses on the application of the low rank adaptation (LoRA) method in training Indonesian automatic speech recognition (ASR) models. This method was chosen for its effectiveness in reducing model size and accelerating training while maintaining high accuracy. The research begins with...
Saved in:
| Main Author: | |
|---|---|
| Format: | Final Project |
| Language: | Indonesia |
| Online Access: | https://digilib.itb.ac.id/gdl/view/75275 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Institut Teknologi Bandung |
| Language: | Indonesia |
| Summary: | This research focuses on the application of the low rank adaptation (LoRA) method in training Indonesian automatic speech recognition (ASR) models. This method was chosen for its effectiveness in reducing model size and accelerating training while maintaining high accuracy. The research begins with an exploration of the fundamental theories related to the LoRA method, ASR technology, and previous research on ASR and LoRA. The study then proceeds with training the model using an Indonesian audio dataset and implementing LoRA techniques during the training process. The results of the research demonstrate a significant improvement in model training speed compared to traditional methods, without sacrificing the accuracy of the word error rate (WER) in the transcription. This study also discusses the challenges encountered and the potential for further research. It is hoped that this research can serve as a stepping stone for further utilization of the LoRA method in the field of speech, particularly for the Indonesian language. |
|---|