App for speech reconstruction -- 3
Budget: $250 – $750 USD
Design and Implement a voice conversion model that works in real-time the input should be Arabic words or sentences in the noisy case and the output should correspond to pure or normal speech.
Examples of input and output attached below,
The steps for it should include:
1- Filtering and noise reduction to enhance the input noisy speech signals.
2- In previous work, I found that GRU is better.
3- The output should be reconstructed using one of the available speech coders, and for me, in previous experiments, I found that ACELP is the best among the available standard speech coders.
4- The code would be written in Python in Google Collab to be used later to implement it as a mobile app.
5- the model evaluation would be beside the recognition rate and WER, subjective test with PESQ and STOI and objective tests like SNR, MSE
Examples of input and output attached below,
The steps for it should include:
1- Filtering and noise reduction to enhance the input noisy speech signals.
2- In previous work, I found that GRU is better.
3- The output should be reconstructed using one of the available speech coders, and for me, in previous experiments, I found that ACELP is the best among the available standard speech coders.
4- The code would be written in Python in Google Collab to be used later to implement it as a mobile app.
5- the model evaluation would be beside the recognition rate and WER, subjective test with PESQ and STOI and objective tests like SNR, MSE