Research map / Large language models / Language, safety and interpretability
HyPoradise: An Open Baseline for Generative Speech Recognition with Large Language Models
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add n in-context demonstrations (0 to 10) (ablation) | GPT-3.5 ICL · WSJ-dev93 | WER (%): 8.5 -> 7.1 | Helped |
| zero-shot prompting on CV-sg | GPT-3.5 ICL · CV-sg | WER (%): 26.4 -> 26.5 | No clear effect |
| use LoRA instead of full fine-tuning (LLaMA) (ablation) | H2T · WSJ | WER (%): 3.8 -> 2.2 | Helped |
| use LoRA instead of full fine-tuning (LLaMA) (ablation) | H2T · SwitchBoard | WER (%): 18.4 -> 14.1 | Helped |
| H2T-ft full fine-tuning vs baseline | H2T-ft · CORAAL | WER (%): 21.4 -> 23.1 | Hurt |
| generative correction with LoRA vs LM rescoring | H2T-LoRA · WSJ | WER (%): 4.3 -> 2.2 | Helped |
| H2T-LoRA vs n-best oracle upper bound for reranking | H2T-LoRA · CHiME-4 | WER (%): 9.1 -> 6.6 | Helped |
| H2T-LoRA vs n-best oracle | H2T-LoRA · CORAAL | WER (%): 21.8 -> 19.2 | Mixed |
| zero-shot LLaMA-13B correction of Whisper hypotheses | LLaMA-13B · CHiME-4 | WER (%): 11.5 +/- 0.5 | No clear effect |
| correct hypotheses with H2T-LoRA before intent detection | H2T-LoRA · ATIS intent detection | accuracy (%): 97.4 vs 97.9 ground truth | Helped |
| T5-large vs LLaMA-13B under LoRA (ablation) | H2T-LoRA · Tedlium-3 | WER (%): 7.4 vs 4.6 | Mixed |
Closest papers by meaning
- Large Language Models are Efficient Learners of Noise-Robust Speech Recognition (ICLR 2024 · similarity 0.91)
- Speech Recognition Meets Large Language Model: Benchmarking, Models, and Exploration (AAAI 2025 · similarity 0.88)
- Recording for Eyes, Not Echoing to Ears: Contextualized Spoken-to-Written Conversion of ASR Transcripts (AAAI 2025 · similarity 0.88)
- It's Never Too Late: Fusing Acoustic Information into Large Language Models for Automatic Speech Recognition (ICLR 2024 · similarity 0.87)
- FastLongSpeech: Enhancing Large Speech-Language Models for Efficient Long-Speech Processing (NeurIPS 2025 · similarity 0.87)
- Audio Large Language Models Can Be Descriptive Speech Quality Evaluators (ICLR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.