e2chndr/evoo-ai-v1
e2chndr/evoo-ai-v1 is a 0.8 billion parameter small language model, fine-tuned from Qwen3-0.6B, specifically designed to polish dictated text. It excels at removing fillers, applying self-corrections, and fixing punctuation and capitalization, transforming raw speech recognition output into clean, typable text. This model is optimized for dictation cleanup, not for answering questions or generating new content, and operates with a 32768-token context length.
Loading preview...
Evoo AI v1: Dictation Polishing Small Language Model
Evoo AI v1 is a compact 0.8 billion parameter language model, fine-tuned from Qwen3-0.6B, with a 32768-token context length. Its primary function is to refine raw speech-to-text output, making it suitable for typing. This model is the polish component of Evoo, an open-source macOS voice dictation application, and was notably fine-tuned entirely on an 8 GB M1 MacBook.
Key Capabilities
- Dictation Polishing: Transforms speech recognizer output by removing fillers, stutters, and self-corrections.
- Punctuation and Capitalization: Automatically corrects and applies appropriate punctuation and capitalization.
- Intent Preservation: Polishes dictated questions or requests without attempting to answer or act on them.
- Performance: Achieves a median word error of 0.14 and 78 "exactly right" results on 300 held-out dictations when used with Evoo's deterministic rules, significantly outperforming stock Qwen3-0.6B.
- Efficiency: Trained using LoRA (rank 8, scale 20) on 16 of 28 layers, resulting in only 2.9M trained parameters (0.48% of the base model).
Good For
- Voice Dictation Applications: Ideal for integrating into systems where raw speech input needs to be cleaned and formatted for text entry.
- Improving Speech-to-Text Accuracy: Enhances the readability and correctness of text generated by speech recognition engines.
- Resource-Constrained Environments: Its small size and efficient training (on an M1 MacBook) make it suitable for on-device or edge deployments.
- Specific Text Transformation: Use cases requiring the transformation of informal spoken language into formal written text, while retaining the original meaning and intent.