Testing different solutions for speech-to-text