vTTS: visual-text to speech

This paper proposes visual-text to speech (vTTS), a method for synthesizing speech from visual text (i.e., text as an image). Conventional TTS converts phonemes or characters into discrete symbols and synthesizes a speech waveform from them, thus losing the visual features that the characters essentially have. Therefore, our method synthesizes speech not from discrete symbols but from visual text. The proposed vTTS extracts visual features with a convolutional neural network and then generates acoustic features with a non-autoregressive model inspired by FastSpeech2. Experimental results show that 1) vTTS is capable of generating speech with naturalness comparable to or better than a conventional TTS, 2) it can transfer emphasis and emotion attributes in visual text to speech without additional labels and architectures, and 3) it can synthesize more natural and intelligible speech from unseen and rare characters than conventional TTS.