update url for file

Author: RCmags

collections/blog/Journal/_posts/2025-03-10-dictation-writting.md

@@ -14,7 +14,7 @@ On my [ThinkPad T540P](https://static.lenovo.com/shop/emea/content/pdf/ThinkPad/
 When combined with other AI tools like Large Language Models (e.g. [Mistral:7b](https://ollama.com/library/mistral) or [DeepSeek-R1:8b](https://ollama.com/library/deepseek-r1) ), these transcribed texts can be refined significantly. For instance, pairing SpeechNote's transcription with a local LLM service like [Ollama](https://ollama.com/) allows the LLM to correct spelling and improve sentence structure. This results in a fairly refined piece of text. Therefore, both tools complement each other well: we transcribe thoughts from speech, then use an LLM model to polish that the transcribed text into something clearer and more readable.
 
 ![image](/diagrams/ai-dictation/ai-speech-correction.png)
-**File**: [ai-speech-correction.drawio](/diagrams/ai-speech-correction.drawio)
+**File**: [ai-speech-correction.drawio](/diagrams/ai-dictation/ai-speech-correction.drawio)
 
 This refined text can be reviewed and edited by the user for final touches, making it closer to their intended expression. Critically, this approach drastically reduces the manual effort required compared to writing from scratch. Moreover, I believe speaking naturally leverages neural pathways that are optimized for rapid idea generation. Speech is our primary mode of communication and it forces us to express clear ideas quickly. Writing might involve over-refining thoughts that would flow more freely during speech. Transcription appears to bypass these internal filters and makes communicating an idea much more natural. The speed increase is substantial: dictation yields far more words per minute than manual typing ever can. It only requires an accurate transcription model and a sufficiently powerful computer.