Enhancing Whisper's Accuracy and Speed for Indian Languages through Prompt-Tuning and Tokenization
This work addresses the challenge of enhancing multilingual speech recognition for low-resource Indian languages, representing an incremental improvement over existing Whisper models.
The paper tackled the problem of Whisper's poor performance in low-resource Indian languages by proposing prompt-tuning with language family information to improve accuracy and a novel tokenizer to reduce inference time, achieving a balance between optimal word error rate and speed across various model sizes.
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.