Demo page for "Multi-speaker Emotional Text-to-speech Synthesizer"

Brief information

• Title: Multi-speaker Emotional Text-to-speech Synthesizer
• Authors: Sungjae Cho, Soo-Young Lee
• Conference: INTERSPEECH 2021
• Session: Show and Tell demonstation
• Presentation time: 16:00-18:00 (GMT+2) & 23:00-01:00 (GMT+9), September 1, 2021
• Compulsory 3-minute video: [Video] [Slides]
• Paper: [PDF] [OfficialPage]
  • Sungjae Cho, and Soo-Young Lee (2021) Multi-Speaker Emotional Text-to-Speech Synthesizer. Proceedings of Interspeech 2021, 2337-2338
  • Cho, S., & Lee, S.-Y. (2021) Multi-Speaker Emotional Text-to-Speech Synthesizer. Proc. Interspeech 2021, 2337-2338
• Paper Abstract: We present a methodology to train our multi-speaker emotional text-to-speech synthesizer that can express speech for 10 speakers' 7 different emotions. All silences from audio samples are removed prior to learning. This results in fast learning by our model. Curriculum learning is applied to train our model efficiently. Our model is first trained with a large single-speaker neutral dataset, and then trained with neutral speech from all speakers. Finally, our model is trained using datasets of emotional speech from all speakers. In each stage, training samples of each speaker-emotion pair have equal probability to appear in mini-batches. Through this procedure, our model can synthesize speech for all targeted speakers and emotions. Our synthesized audio sets are available on our web page.

Brief introduction to demo

In the following demonstration, audios synthesized by our multi-speaker emotional text-to-speech synthesizer are demonstrated. Our synthesizer can synthesize speech for 7 emotions (neutral, angry, disgust, fear, happiness, sadness, and surprise) across 10 speakers (5 females, 5 males).

We, humans, recognize emotion from speech, depending on what textual content is contained. Thus, for each emotion, we present 3 sentences containing content of a particular emotion. For neutral emotion, we synthesized 3 sentences for all possible emotion-speaker pairs. For another emotion, we synthesized 3 sentences for neutral and that textual emotion across all speakers.

In the first demonstration, synthesized utterances of neutral sentences are presented. What we want you to pay attention to is how varying given audios are across emotions and speakers. We recommend you to listen to a neutral audio, and then listen to an emotional audio. Next, check out what varies across speakers. Note that the blue boxes where audios were synthesized even without training supervision. Those expressions were transferred from other speakers' expressions.

You can move onto another page to listen to audios for another emotional sentence, through the following hyperlinks.

[Neutral sentences (home)] [Angry sentences] [Disgusted sentences] [Fearful sentences] [Happy sentences] [Sad sentences] [Surprised sentences]

[One page view]

Neutral sentences

Neutral sentence 1

Neutral sentence 2

Neutral sentence 3

Acknowledgement

This work was supported by Ministry of Culture, Sports and Tourism and Korea Creative Content Agency [R2019020013, R2020040298].