There are more than 70 million people worldwide who suffer from stuttering problems. This will affect the confidence of public speaking in people who suffer from this issue. To solve this problem many people take therapy sessions but the therapy sessions are a temporary solution, as soon as they leave therapy sessions this problem might arise again. This work aims to use state of the art machine learning algorithms that have improved over the past few years to solve this problem. We have used the dataset from UCLASS archives which provide the data for stuttered speech in.wav format with time-aligned transcriptions. We have tried different algorithms and optimized our model by hyper parameter tuning to maximize the model’s accuracy. The algorithm is tested on random speech data with low to heavy stuttering from the same dataset, and it is observed that there is significant reduction in the Word Error Rate (WER) for most of the test cases.