Speech Recognition Using RNN, DNN and Web Services

Human-Computer Interactions (HCI) are extremely important in today’s fast developing AI environment. The process of voice recognition involves taking the most important data from the incoming speech signal and using that information to determine accurately what text goes where. As Alexa and Siri only translate foreign languages, our goal is to convert the speech signal for eight Indian languages into text. User speaks through microphone during speech recognition system. In order to discriminate between various words and sounds, characteristics are taken from the spectrogram. Mel Frequency Cepstral Coefficients are the procedure utilised for Feature Extraction (MFCC). A probability over alphabetic characters is then generated by the Acoustic Model using the retrieved features as input. The collected features are mapped to words and sounds using DNN and RNN algorithms in this step. In order for the algorithms to learn the connections between the features and the sounds they represent, a huge corpus of speech data is used for training. By supplying data on word relationships within a language and the statistical likelihood of word sequences, language models play a key role in speech recognition systems. The likelihood of a word sequence is here estimated using probabilistic models using the audio that has been transcribed. It is possible to develop language models using statistical methods, such as n-gram language models and recurrent neural network language models (RNN-LMs). A suitable class label is applied to a pattern based on an abstraction created using a collection of training patterns or domain expertise, and the pattern is then converted. We provide results on the recognition of non-native speech using multilingual Hidden Markov Models. The system’s effectiveness and optimisation are based on how long it takes to translate into the target language. Finally, the services are available in the web so that anyone can access and perform Speech-to-text conversion.