Offline continuous speech recognition for mobile devices

Speech recognition is an option of accessibility in electronic devices. Currently, this recognition is accomplished through Application Programming Interfaces that depend on Internet connection and which are often made available by proprietary software. Voice recognition is an important technology for improving HCI (Human-Computer Interaction), after all speech is a human feature that most people have. The increased use of adaptive interfaces using speech recognition results from the fact that speech is the most natural form of interaction. Speech makes it faster to access information in software and applications, compared to standard interaction forms such as touchscreen, mouse, keyboard, among others. Considering this context, this work presents an open-source solution for continuous speech recognition in mobile devices. At first, a tool comparing the processing and memory usage of library configurations on a desktop computer was developed. Then, we implemented the best alternatives in an Android application and tested its performance and battery usage on several mobile devices. For this, the best libraries were selected, and their configuration files were optimized to find the most cost-effective solution between performance and accuracy of the Word Error Rate (WER) and Sentence Error Rate (SER) rates.