WhisperNow releases locally fine-tuned transcription cleanup models
Developer stephan273 released WhisperNow, an MIT-licensed voice-to-text app for Windows, macOS, and Linux, together with three locally fine-tuned models, a 66,557-row dataset, and training documentation. The LFM2.5 1.2B, LFM2.5 2.6B, and Qwen3.5 4B models were trained to remove fillers, false starts, and self-corrections and to format lists locally. The author reports that they outperformed several cloud models in factual-error tests, although those performance results have not been independently verified.
WhisperNow provides an open-source, locally run alternative for polishing dictated text, with model weights, training data, and methodology released alongside the application.
#WhisperNow#LFM2.5 1.2B#LFM2.5 2.6B#Qwen3.5 4B#gpt-6.1-sol#Claude Haiku 5.5#gpt-6-luna#glm-5.3-flash#glm-5.3#Modal