Summary of Improving Domain-specific Asr with Llm-generated Contextual Descriptions, by Jiwon Suh et al.
Improving Domain-Specific ASR with LLM-Generated Contextual Descriptions
by Jiwon Suh, Injae Na, Woohwan Jung
First submitted to arxiv on: 25 Jul 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper proposes a novel approach to improve end-to-end automatic speech recognition (E2E ASR) systems for recognizing domain-specific words. It leverages the state-of-the-art Whisper model without modifying its architecture, while enabling it to utilize descriptions effectively. The method also incorporates two additional training techniques: decoder fine-tuning and context perturbation. Furthermore, the paper suggests using a Large Language Model (LLM) to generate descriptions with simple metadata when actual descriptions are unavailable. Experimental results demonstrate that these methods significantly enhance domain-specific ASR accuracy on real-life datasets. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper tries to make speech recognition better by helping machines understand specific words and terms. It uses a powerful model called Whisper without changing its original design, allowing it to use explanations effectively. The method also includes two new ways to train the model: fine-tuning and shaking things up with context. Additionally, it proposes using another large language model to create simple descriptions when real ones aren’t available. By doing this, the methods show that they can greatly improve how well machines recognize specific words on real-life recordings. |
Keywords
* Artificial intelligence * Decoder * Fine tuning * Large language model
