Aye10032/Qwen3-ASR-Refiner-1.7B
Aye10032/Qwen3-ASR-Refiner-1.7B is a 1.7 billion parameter model from the Qwen3 family, specifically fine-tuned to convert spoken-style Chinese ASR transcripts into formal, natural written Chinese. Developed by Aye10032, this model excels at refining text while preserving original meaning and not introducing new information. It is optimized for Chinese speech-to-text post-processing applications, ensuring high-quality written output from conversational input.
Loading preview...
Qwen3 ASR Refiner: Chinese Spoken-to-Written Text Conversion
This model, Aye10032/Qwen3-ASR-Refiner-1.7B, is part of a family of models developed by Aye10032, specifically designed for refining Chinese ASR (Automatic Speech Recognition) transcripts. Built upon the Qwen3-1.7B base model, its primary function is to transform informal, spoken-style Chinese text into concise, natural, and formal written Chinese.
Key Capabilities
- Spoken-to-Written Conversion: Converts conversational Chinese into a more formal written style.
- Meaning Preservation: Ensures the original meaning of the text is maintained without adding new information.
- Concise Output: Generates refined text that is both natural and to the point.
- Fine-tuned for Chinese: Optimized specifically for the nuances of Chinese language transformation.
Training and Architecture
All variants in the Qwen3 ASR Refiner family are fine-tuned using the Aye10032/WenetSpeech-Formal-Text dataset. The LoRA adapter used during fine-tuning has been merged into the base model, allowing for direct loading of the complete BF16 Transformers weights without requiring PEFT (Parameter-Efficient Fine-Tuning) libraries. This 1.7 billion parameter model offers a robust solution for enhancing the quality of Chinese ASR outputs.