speech recognition platform

I would like to offer you a job that involves developing a speech engine for urdu language. I can share details latter if you are interested. details are as follows; Training Acoustic Model (AM) Goal: Train Urdu AM by hybrid HMM (Hidden Markov Model) & DNN Steps and Deliverables: - Analysis acoustic feature: calculate MFCC (Mel Frequency Cepstrum) and do CMVN (Cepstrum Mean Variance Normalization) for noise-robust. (feature vectors files) - VAD: Voice Activity Detection(DNN based VAD or other).(VAD model and vad result) - Train HMM AM and align feature vectors. (mono-phone HMM and Triphone HMM dictionary) - Train LDA (Linear Discriminate Analysis) and MLLT(Maximum Likelihood Linear Transform) matrices.(transform matrices) - Train fMLLR (frame-based Maximum Likelihood Linear Regression) and SAT (Speaker Adaptation Transform) matrices. (transform matrices) - Next, get the alignment for DNN AM training. (basic data for DNN training) - Train TDNN (Time Delay NN) & LSTM (Long short-term memory) NN.(DNN model file) In this step, we can test various DNN architectures for the best performance. Decoding Urdu speech by the trained DNN AM and LM. Goal: Construct Urdu speech recognition system by the trained DNN and test its performance. Steps and Deliverables: - Read test wave file and analysis MFCC features. - VAD and CMN - search the best path on lattice. - evaluate the performance of Urdu STT. Budget: $1000 The deadline is 2 weeks max and expected accuracy is above 90%