Pseudo-Labeling for Domain-Agnostic Bangla Automatic Speech Recognition
Authors:
- Rabindra Nath NandiRNRabindra Nath Nandi
- Tariqul IslamTITariqul Islam
- Shammur Absar ChowdhurySCShammur Absar Chowdhury
Preprint
Abstract
One of the major challenges for developing automatic speech recognition (ASR) for low-resource languages is the limited access to labeled data with domain-specific variations. In this study, we propose a pseudo-labeling approach to develop a large-scale domain-agnostic ASR dataset. With the proposed methodology, we developed a 20k+ hours labeled Bangla speech dataset covering diverse topics, speaking styles, dialects, noisy environments, and conversational scenarios. We then exploited the developed corpus to design a conformer-based ASR system. We benchmarked the trained ASR with publicly available datasets and compared it with other available models. To investigate the efficacy, we designed and developed a human-annotated domain-agnostic test set composed of news, telephony, and conversational data among others. Our results demonstrate the efficacy of the model trained on psuedo-label data for the designed test-set along with publicly-available Bangla datasets. The experimental resources will be publicly available.(https://github.com/hishab-nlp/Pseudo-Labeling-for-Domain-Agnostic-Bangla-ASR)Keywords
All Related Versions
This publication has 10 references indexed in Scilit:
- Fast Conformer With Linearly Scalable Attention For Efficient Speech Recognition2023
- Alternative Pseudo-Labeling for Semi-Supervised Automatic Speech RecognitionIEEE/ACM Transactions on Audio, Speech, and Language Processing, 2023
- Momentum Pseudo-Labeling: Semi-Supervised ASR With Continuously Improving Pseudo-LabelsIEEE Journal of Selected Topics in Signal Processing, 2022
- Bangladeshi Bangla speech corpus for automatic speech recognition researchSpeech Communication, 2022
- slimIPL: Language-Model-Free Iterative Pseudo-Labeling2021
- Towards One Model to Rule All: Multilingual Strategy for Dialectal Code-Switching Arabic ASR2021
- Developing RNN-T Models Surpassing High-Performance Hybrid Models with Customization Capability2020
- Conformer: Convolution-augmented Transformer for Speech Recognition2020
- Improving ASR Systems for Children with Autism and Language Impairment Using Domain-Focused DNN Transfer Techniques2019
- Semi-Orthogonal Low-Rank Matrix Factorization for Deep Neural Networks2018