This open-source ASR solution is based on the open-source project, Kaldi. Kaldi is a great toolset leveraging state of the art ASR algorithms. But in order to use it in production, you need two important things:
- An ASR model with hundreds or even thousands of hours of transcribed audio data and months of fine tuning.
- Scripts to run the model, and a FastCGI server which will allow you to do server side voice recognition over HTTP protocol (Apache, Nginx, Microsoft IIS, etc.).
API.AI is openly sharing our ASR model under the Creative Commons Attribution-ShareAlike 4.0 International Public License so that you could get decent and accurate speech recognition out of the box. They are also open sourcing the scripts required to get ASR up and running on your server.