Towards Artificial Intelligence-Generated Texts as a Vocal Loading Task: Can They Replace Standard Texts?
Related Experiment Video
Updated: Nov 5, 2026
Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness
Published on: December 6, 2024
Saeed Saeedi1, Mehdi Bakhtiar1
1Speech and Neuromodulation Laboratory, Unit of Human Communication, Learning and Development, Faculty of Education, The University of Hong Kong, Hong Kong SAR, China.
Objectives:
The problem with standard available texts as vocal loading tasks is that they might not be long enough to induce vocal fatigue. On the other hand, finding texts from other sources can be time-consuming, making it hard to compare outcomes across different studies. This study sought to create a set of texts for vocal loading tasks and evaluate their readability and linguistic features.
Methods:
The Google Gemini artificial intelligence (AI) chatbot was prompted to create simple texts about daily topics while keeping the sentences short and simple in a way that would be readable for a sixth-grade audience. A total of 20 texts were generated. Readability via Flesch Reading Ease Score (FRES), Gunning Fog Index (GFI), Flesch-Kincaid Grade Level (FKGL), and Dale-Chall Raw Score (D-CS), while language proficiency level via the Common European Framework of Reference for Languages (CEFR) were examined for each text. Additionally, eight traditional standard texts were analyzed to provide a comparative baseline.
Results:
AI-generated texts had a mean of 419.85 ± 29.95 words and 31.40 ± 1.69 sentences. The readability indices satisfied the criteria for a sixth-grade or lower reading level, with an FRES of 92.91 (≥ 80.0), GFI of 6.34 (≤ 6.9), FKGL of 3.63 (≤ 6.9), and D-CS of 1.47 (≤ 5.9). Most of the vocabulary (94.07%) in the generated texts matched what is required for basic users of English. In terms of the distribution of phonemes, the texts consisted of 21.20% voiceless consonants, 38.35% voiced consonants, and 40.45% vowels.
Conclusions:
The AI-generated texts consistently achieved a readability level at or below the sixth-grade, and they were found to be compatible with a basic level of English proficiency. These preliminary readability and linguistic insights indicate that these texts can serve as a practical and standardized set of materials for vocal loading tasks. However, further empirical research with human participants is warranted to validate their physiological efficacy to induce vocal fatigue.
