S1 ASR Cleanup
No description provided
About This Project
S1-ASR-Cleanup is a specialized fine-tune of the Qwen3-0.6B-based S1-mini model, optimized for cleaning up transcriptions generated by speech-to-text systems. It intelligently processes raw dictation by handling complex tasks such as paragraph segmentation, currency normalization, date formatting, and the expansion of informal contractions. The model also resolves common speech disfluencies, removes self-corrections, and corrects mid-sentence capitalization errors to produce a final, structured text output.
Built using the transformers library, this model is designed as a standalone solution with the LoRA merged directly into the base model. It is particularly useful for users who rely on voice dictation for daily productivity and require a fast, deterministic way to convert messy audio transcripts into clear, professional-grade documentation.