Publication

Abstract : Text-To-Speech Synthesis (TTS) is an effective research area which produces synthetic speech from underlying text. To mimic genuine speech, the detected syllables are delivered with the appropriate duration and prosodic traits. This proposed system's goal is to synthesize human parity with natural sound for the input domain text in English language. The primary output must be a synthetic natural human sounded audio signal. Artificial Intelligence algorithms and various exploratory techniques are leveraged to process text and synthesis human speech in real-time. The effectiveness of synthetic speech has substantially increased by using neural network-based end-to-end TTS, it makes the speech sound natural and more human sounding. Existing state of art TTS techniques are explored, various pretrained models are trained and the quality of synthetic voice is assessed. Open-source avionics related textual corpus is used to train and assess the system. All the models are applied to a variety of audio generation tasks. Spectrograms of generated audios were compared to that of human generated audio spectrograms to assess how system generated spectrograms could be developed further to improve the quality of the synthetic voice. © 2023 IEEE.