Harf-Speech: A Clinically Aligned Framework for Arabic Phoneme-Level Speech Assessment
Automated phoneme-level pronunciation assessment plays a crucial role in enabling scalable speech therapy and language learning solutions. However, validated tools specifically designed for the Arabic language have been limited, hindering progress in this essential area. In response to this challenge, a new system known as Harf-Speech has been developed to provide a comprehensive framework for assessing Arabic pronunciation at the phoneme level.
Harf-Speech combines several advanced technologies and methodologies to achieve its goal. The system features a modular architecture that integrates the following components:
- MSA Phonetizer: Converts Modern Standard Arabic (MSA) text into phonetic representations.
- Fine-tuned Speech-to-Phoneme Model: Analyzes spoken Arabic and translates it into phoneme sequences.
- Levenshtein Alignment: Utilizes string metrics to align phonemes accurately.
- Blended Scorer: Combines longest common subsequence and edit-distance metrics to evaluate pronunciation accuracy.
To ensure the system’s effectiveness, the developers fine-tuned three Automatic Speech Recognition (ASR) architectures on a diverse set of Arabic phoneme data. These models were then benchmarked against zero-shot multimodal models to determine their performance. Among these, the OmniASR-CTC-1B-v2 model stood out, achieving an impressive phoneme error rate of just 8.92%.
To validate the clinical relevance of Harf-Speech, three certified speech-language pathologists independently evaluated a selection of 40 utterances. The results of this evaluation revealed that Harf-Speech achieved a Pearson correlation of 0.791 and an Intraclass Correlation Coefficient (ICC(2,1)) of 0.659 when compared to the mean expert scores. These metrics indicate that Harf-Speech significantly outperforms existing end-to-end assessment frameworks in terms of clinical alignment and interpretability.
The implications of Harf-Speech are profound for both speech therapy and language learning in Arabic-speaking populations. By providing a clinically validated, phoneme-level pronunciation assessment tool, Harf-Speech promises to enhance the effectiveness and accessibility of speech therapy. Additionally, language learners can benefit from precise feedback on their pronunciation, enabling them to improve their skills more effectively.
In conclusion, Harf-Speech represents a significant advancement in the field of speech assessment for Arabic. Its innovative approach, combining cutting-edge technology and clinical expertise, positions it as a valuable resource for professionals and learners alike. As research continues, the potential for further enhancements and applications of Harf-Speech may pave the way for even greater breakthroughs in the realm of speech therapy and language education.