Automated speech therapy through personalized pronunciation correction using reinforcement learning and large language models
Bibliographic record
Abstract
Traditional approaches to pronunciation correction often face challenges in personalization, adaptability, and consistent feedback. This study introduces a novel AI-powered system that integrates Reinforcement Learning (RL) and Large Language Models (LLMs) to address these limitations. The system employs a custom Proximal Policy Optimization (PPO) algorithm for precise pronunciation evaluation and an Large Language Models to deliver detailed, encouraging, and user-specific feedback. It was evaluated using the CMU Sphinx Dictionary dataset, a foundational phonetic resource, alongside dynamically generated user-specific session data for personalized feedback and model refinement. Further validation utilized datasets such as TIMIT, LibriTTS, SpeechOcean762, and the Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS), enabling direct comparisons with contemporary methods. Results demonstrate the system's robustness in handling diverse phonetic variations. While primarily tested on English data, its modular architecture supports adaptation to other languages and dialects through language-specific phonetic datasets. The system achieved exceptional performance metrics: 97.9 % phoneme-level accuracy, 87.7 % word-level accuracy, 95.2 % syllable count accuracy, and 89.4 % perfect accuracy on the CMU Sphinx dataset. This innovative approach underscores the potential of advanced AI techniques to enhance the personalization and effectiveness of pronunciation correction systems. All findings are quantitatively validated and thoroughly documented.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.003 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.004 | 0.003 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".