Enhancing Whisper's Accuracy and Speed for Indian Languages through Prompt-Tuning and Tokenization

TL;DR

This paper improves Whisper's speech recognition accuracy and speed for Indian languages by using prompt-tuning with language family info and a new tokenizer, balancing accuracy and efficiency.

Contribution

It introduces prompt-tuning with language family data and a novel tokenizer to enhance multilingual speech recognition in low-resource Indian languages.

Findings

Tokenizer reduces inference time significantly.

Prompt-tuning improves accuracy across model sizes.

Combined methods balance WER and speed.

Abstract

Automatic speech recognition has recently seen a significant advancement with large foundational models such as Whisper. However, these models often struggle to perform well in low-resource languages, such as Indian languages. This paper explores two novel approaches to enhance Whisper's multilingual speech recognition performance in Indian languages. First, we propose prompt-tuning with language family information, which enhances Whisper's accuracy in linguistically similar languages. Second, we introduce a novel tokenizer that reduces the number of generated tokens, thereby accelerating Whisper's inference speed. Our extensive experiments demonstrate that the tokenizer significantly reduces inference time, while prompt-tuning enhances accuracy across various Whisper model sizes, including Small, Medium, and Large. Together, these techniques achieve a balance between optimal WER and inference speed.