A DNN-BASED REAL-TIME SPEECH-TO-TEXT CONVERSION

The instant voice or recorded speech is converted to text format in real-time concurrent speech-to-text systems.As there are multiple ways of doing this operation, we challenge ourselves to verify the WER of the system. We created a real-time speech-to-text conversion system using the DNN techniques in the GOOGLE COLLABORATORY platform which is trained with the LJ Speech databases that contain the transcription of every speech to verify the word error rate. DNN has been proven to be the better technique for predicting the system’s desired output. This speech recognition model, which we refer to as DeepSpeech 2, was created by combining 2D CNN, RNN, and CTC Loss.Due to this, the model is getting a lower Word Error Rate and loss.