Sustained Vowel Game: A Computer Therapy Game for Children with Dysphonia

Vanessa Lopes, João Magalhães, Sofia CavacoInterspeech 2019

TL;DR

This paper introduces the "Sustained Vowel Game," a novel computer-based therapy tool designed to make vocal exercises more engaging for children with dysphonia. By incorporating visual feedback and gamified elements, the game aims to improve children's motivation and performance in sustained vowel exercises, a crucial component of voice therapy.

What Problem Does This Paper Address?

Children aged 4-12 can experience dysphonia, a voice disorder that impacts communication, social interaction, and self-esteem. A common therapeutic intervention for dysphonia is the sustained vowel exercise (SVE), which involves holding a vowel sound for an extended period with stable pitch and loudness. However, for children, this exercise can become monotonous and tedious, leading to reduced adherence and effectiveness. Existing computer-based therapy tools for speech and language disorders often lack specific applications for voice therapy, and those that do exist may not be sufficiently engaging for young children. This paper addresses the need for a more motivating and interactive SVE therapy tool for children with dysphonia.

Methods

The core of the "Sustained Vowel Game" relies on an automatic vowel recognition system to provide real-time feedback. This system uses Mel-frequency cepstral coefficients (MFCCs) as features, extracted from audio samples. A Support Vector Machine (SVM) with a radial basis function (RBF) kernel was chosen as the classification algorithm due to its high accuracy. The game itself incorporates several key features:

  • Gamified Interface: Attractive scenarios and characters are used to motivate children.
  • Visual Feedback: The movement of a character on screen provides immediate visual cues about the child's performance. If the voice intensity deviates from the target, the character's position changes, helping the child understand how to correct their production.
  • Parameter Control: Speech-language pathologists (SLPs) can customize parameters such as target intensity level and maximum phonation time (MPT) for each child.
  • Dynamic Difficulty Adjustment: The game aims to keep children engaged by dynamically adjusting the challenge level.

Datasets

The vowel classification model was trained and validated using a combination of two European Portuguese (EP) vowel datasets:

  • Ferreira's vowel dataset: Includes the phonemes /a/, /ε/, /i/, /ɔ/, /u/.
  • BioVisualSpeech's vowel dataset: Includes the sustained vowels /a/, /i/, and /u/.

These datasets were combined to create a dataset from 48 children aged 4-10 years old.

Key Insights and Ideas

  • The Power of Visual Feedback: The paper highlights how simple visual cues, like a character falling when voice intensity drops, can significantly help children understand and correct their vocal productions in real-time. This is a crucial insight for designing effective child-computer interaction in therapy.
  • Gamification for Adherence: The study strongly suggests that gamified elements, such as rewards and engaging scenarios, are effective in increasing children's motivation to practice SVEs, a task that can otherwise be perceived as boring.
  • MFCCs Remain a Strong Feature: Despite exploring various feature sets, MFCCs (specifically 12-dimensional vectors) combined with an SVM proved to be highly effective for accurate vowel classification in this context.
  • Time-Domain Information is Crucial: The results indicated that feature vectors that retain temporal information (like MFCCs) perform better than those that only capture statistical summaries (mean and standard deviation) of acoustic features.

Key Results

A user test with 14 children (including three with diagnosed dysphonia) yielded promising results:

  • Increased Phonation Time: Children using the game achieved significantly longer maximum phonation times (MPT) compared to when performing the SVE without the game. For example, participants 10 and 12 showed an improvement of 8 seconds from trial 1 to trial 3.
  • Motivation and Engagement: All children, including those with dysphonia, showed increased interest and motivation when using the game compared to the non-game trial. They expressed a desire to continue playing to earn rewards and explore different scenarios.
  • Dysphonia Improvement: While children with dysphonia faced more challenges, two of them managed to achieve MPTs close to the normative value lower bound. Even a child with asthenia, who initially struggled with adjusted MPT settings, was able to complete the task successfully after parameter adjustments, demonstrating the game's adaptability.
  • High Accuracy Model: The chosen vowel classification model, based on SVM and 12 MFCCs, achieved a high accuracy score of 96%.

Limitations and Future Work

  • Limited Dysphonia Scope: The user test included only three children with diagnosed dysphonia. Further studies with a larger and more diverse group of children with different types of dysphonia would be beneficial.
  • Parameter Setting: While SLPs can parameterize the game, the paper doesn't delve deeply into the optimal strategies for setting these parameters for different children and pathologies.
  • Long-Term Effects: The study focused on short-term improvements in MPT and motivation. Future work could investigate the long-term impact of the game on vocal quality and overall therapeutic outcomes.
  • Broader Vocal Exercises: The current game focuses on sustained vowel exercises. Expanding the game to incorporate other vocal exercises relevant to dysphonia therapy could enhance its utility.

Summary updated

BibTeX
@inproceedings{lopes2019sustained,
  author = {Vanessa Lopes and João Magalhães and Sofia Cavaco},
  title = {{Sustained Vowel Game: A Computer Therapy Game for Children with Dysphonia}},
  booktitle = {Interspeech},
  year = {2019},
}
Figure