Open Internet by MindsNet

Fine-tuning Whisper for Domain-Specific Vocabulary

There is a need for effective methods to fine-tune Whisper, a speech recognition model, on domain-specific vocabulary, particularly in languages like Spanish. The challenge involves determining the required amount of labeled audio for reliable detection of specific words and technical terms. Current methods like LoRA, QLoRA, and Spectrum are known, but there may be newer or better approaches.

Computing & Technology, Computer Science, Machine Learning