Hi there,
You want a fully working system built from scratch without relying on third parties and you need it as soon as possible; this is even difficult for experienced people. I had my Master thesis on face recognition (especially age estimation) which is related to speech recognition and I can say that the best performing architectures till now in both fields are deep neural networks.
If you allow for enough time, I can build your speech recognizer with satisfying accuracy (it well rely on opensource projects tough, but which are free for commercial use).
Please feel free to contact me anytime.
Best regards,
Houssam