Agbeli/Speech_recognition_AMMI

Speech Recognition for Low Resource Language(Twi) # Speech_recognition_AMMI Speech Recognition for Low Resource Language(Twi) - This was an in-class project on speech recognition for low-resource language such as Twi my local dialect. The repos is made up of the following: - Data collected using an app called ligaikuma. - Writtten text of the holy scriptures in our local dialect. - Texts with their corresponding recordings - Two hours of data audio recordings. The calculate the total recordings. - The next phase of the project is to model the data using CPC model. - We built Automatic Speech Recognition (ASR) system by using a pretrained model with Connectionist Temporal Classification on top of it. - We had a Character Error Rate of 0.98 and 0.908 for both training and validation set respectively notebook. ## License: - Under the license of MIT