Download and Extract model under models folder.Download the project code on your machine.2021.4Ĭurrently, you need to build the python environment to run and use this program 😓 For more knowledge about offline STT model, check vosk models for details and stt/vosk_api/README.md local file for how to integrate offline model to this project. However, the offline model does not perform very well. Speech to text transition based on vosk api is implemented and now you can get subtitle files with recognized texts in it.Those will take a long time so I may say this project is not abandoned but stagnated techniquely. I need to review the network and I may choose another network for ms-precision. After updating the model with a much bigger dataset, the model inference result under most of the circumstances still can only parse sentences on second-wise, which lacks precision for practical usage.See TODO.md in the project folder for details. There is still much to improve in real time cases. Be patient to see when it's fully loaded 😄) (In case you don't know why I put the GIF here, the small 'xxx' overlapping are from the output subtitle file generated using this project which shows it konws when people talk. The project is in a primitive phase 😅 Further development will come when I find more time after The Lunar New Year. This project focuses on how to automatically generate subtitle files from an input audio/video file containing all the timelines for dialogues and human speeches.

Auto generate srt file