Special Sessions

SS – Research Advances in Child Speech Technologies

Organizers: Peter Corcoran (University of Galway, Ireland) and Rishabh Jain (University of Galway, Ireland)

Objectives:
  • Discuss applications of speech synthesis and data augmentation in order to improve ASR/TTS technologies for child.
  • Discuss the limitation of traditional supervised learning approaches for low resource child speech language and its workaround using self-supervised and unsupervised learning approaches for improving the child ASR.
  • TTS driven augmentation using transfer learning and fine-tuning methodologies for building synthetic child speech datasets.
  • Challenges and discussion towards audio-visual facial animation for child.

 

Session overview:

Deep learning research in speech technology is improving rapidly. We have models with billions of parameters trained of thousands of hours of data. However, we still reach a still point when it comes to low resource languages such as child speech where data scarcity becomes an issue. TTS technologies can generate near human-like speech, however when it comes to child speech, progress has been quite slow. There is a lack of solution to this domain which comes with the fact about data availability in this area of research. Due to inherent differences between adult and children’s voices, a child’s voice is quite different from an adult voice in terms of pitch, linguistic and acoustic features, ability to understand and pronounce words, high fundamental frequency, and shorter vocal track length. With this session, we would like to bring together researchers to discuss key technologies, their limitations, and developing solutions to improve speech and vision technologies around child speech.

 

Topics of interest (but not limited to):
  • Speech augmentation for improving ASR/TTS for child speech.
  • Child speech recognition as a low resource area.
  • Use of synthetic speech tools and methodologies for improving the child speech technologies.
  • Audio-visual facial animation for child.

________________________________________

Notes:

  • The Special Session follows the same guidelines as the “SpeD 2023” main conference (see the Paper Submission page).
  • On the Submission platform, please make sure you have checked “Special Session on Child Speech Technologies”.