A Lightweight and Efficient Text-to-Speech Framework for Educational Applications

Abstract

Text-to-Speech (TTS) systems play a crucial role in modern e-learning and assistive technologies by facilitating effective human-computer interaction. This paper discusses the creation of an AI-based multilingual TTS framework using Flask and MySQL, which transforms textual input into natural-sounding speech via Google Text-to-Speech (gTTS). The system features intelligent modules, including user authentication, file-based text extraction from formats such as PDF, DOCX, and TXT, automatic text summarisation, and multilingual translation. Its architecture comprises modules for text preprocessing, language mapping, translation handling, neural speech synthesis, and audio generation. Supporting English, Hindi, and Telugu, the system enhances accessibility for diverse learners. Experimental validation showed efficient real-time voice generation with reduced latency, improving usability in educational settings. By combining translation and summarisation with speech synthesis, this framework boosts digital content accessibility and offers a scalable solution for interactive learning. Future developments aim to integrate advanced neural TTS models for enhanced expressiveness and support offline deployment.

Introduction

The text discusses the development of an AI-based multilingual Text-to-Speech (TTS) system designed to improve accessibility and learning in digital educational environments. Traditional TTS systems often produced robotic and unnatural speech, while modern systems enhanced by Artificial Intelligence (AI) and Natural Language Processing (NLP) now provide more natural speech output. However, many existing systems still lack integrated features such as multilingual translation, document text extraction, summarisation, and secure user management. To address these limitations, the proposed framework combines these functionalities into a single lightweight and efficient platform using the Flask framework, MySQL database, and Google Text-to-Speech (gTTS).

The system is motivated by the growing need for accessible educational technologies for visually impaired users, language learners, and auditory learners. It supports text extraction from multiple document formats such as PDF, DOCX, and TXT files, enabling users to upload educational materials directly without manual typing. The extracted text undergoes preprocessing, cleaning, and optional summarisation to improve readability and reduce cognitive load before speech generation.

The framework also includes multilingual translation support, currently supporting English, Hindi, and Telugu. If the selected output language differs from the input language, the system automatically translates the content before generating speech. The speech synthesis module uses gTTS to generate clear and natural MP3 audio files in real time with low latency, making the system suitable for interactive educational applications.

The literature survey highlights major advancements in neural TTS technologies such as Tacotron 2, FastSpeech 2, PromptTTS, and lightweight architectures like LEF-TTS. These systems improved speech naturalness, expressiveness, and efficiency, but most focus only on speech synthesis and do not integrate translation, document processing, summarisation, and secure backend services. This research gap motivated the development of a comprehensive educational TTS framework.

The proposed methodology follows a modular client-server architecture using Flask. The workflow includes user authentication, text input or document upload, text extraction, preprocessing, summarisation, multilingual translation, speech synthesis, and audio delivery. The backend integrates MySQL for secure user registration, login authentication, password validation, and history tracking. This modular structure improves scalability, maintainability, and reliability.

The evaluation results show that the system performs effectively across multiple functions. File uploads and text extraction operate accurately, summarisation generates concise outputs, multilingual translation works successfully, and speech synthesis produces intelligible and natural audio. The system demonstrates low processing latency, stable performance, and efficient operation even on moderate hardware. Additional features such as voice preview, audio download, and a search-related data retrieval module further enhance usability and educational support.

Compared to traditional rule-based TTS systems, the proposed framework provides integrated document extraction, multilingual translation, summarisation, secure authentication, and real-time speech generation in a single platform. Although advanced neural TTS systems may offer more expressive speech, they require higher computational resources. The proposed system achieves a balance between functionality, efficiency, and lightweight deployment, making it suitable for educational and assistive applications.

The discussion concludes that the system successfully validates its design objectives by improving accessibility, reducing manual effort, and supporting intelligent educational content delivery. However, limitations remain, including dependency on internet connectivity for external TTS services, limited language support, and lack of emotional speech modulation. Despite these limitations, the framework effectively bridges the gap between advanced neural TTS research and practical educational deployment, offering strong potential for future improvements and wider real-world applications.

Conclusion

This paper discusses the design and implementation of an AI-based Multilingual Text-to-Speech framework aimed at educational applications. The system incorporates various processes such as document text extraction, preprocessing, translation, summarization, and speech synthesis, which are organized within a lightweight client-server architecture. Evaluation results indicate that the framework produces natural and intelligible speech output while maintaining real-time performance and usability. The incorporation of multilingual support and secure backend management improves accessibility and reliability. The framework is presented as a practical and scalable solution for intelligent speech-enabled learning environments. Future research will focus on integrating advanced neural TTS models, enhancing semantic summarization techniques, introducing emotional speech synthesis, and enabling offline deployment capabilities.

References

[1] Subha P, Prabavathi R, Brindiia Devi V, Vijayalakshmi S, Mohanapriya M, Akshaya V,”AI-based Multilingual Text-to-Video and Speech”, Published in: 2025 7th International Conference on Innovative Data Communication Technologies and Application (ICIDCA), Date of Conference: 06-08 October 2025, DOI: 10.1109/ICIDCA66325.2025.11280562,Date Added to IEEE Xplore: 16 December 2025. [2] Viraj Walavalkar, Nishant Desale, Yashraj Dhole, Janvi Sawalkar, Pranjal Pandit, Anuradha Yenkikar, “AI-Driven Audiobook Production: Advancements, Challenges, And Future Directions”, Published in: 2025 IEEE Pune Section International Conference (PuneCon), Date of Conference: 12-14 December 2025, DOI: 10.1109/PuneCon67554.2025.11379325, Date Added to IEEE Xplore: 17 February 2026. [3] Showrajit Saha, Nursadul Mamun, “CTS-Synthesizer: Handwritten Character to Speech Conversion using CNN and Google Text-to-Speech Synthesizer”, Published in: 2024 27th International Conference on Computer and Information Technology (ICCIT), Date of Conference: 20-22 December 2024, Date Added to IEEE Xplore: 10 June 2025. [4] Youngdo Ahn, Jongwook Chae, Jong Won Shin, “Text-to-Speech With Lip Synchronisation Based on Speech-Assisted Text-to-Video Alignmentand Masked Unit Prediction”, Published in: IEEE Signal Processing Letters ( Volume: 32), Page(s): 961 – 965, DOI: 10.1109/LSP.2025.3537949, Date of Publication: 03 February 2025. [5] Zehai Tu, Guangyan Zhang, Yiting Lu, Adaeze Adigwe, Simon King, Yiwen Guo, “Enabling Beam Search for Language Model-Based Text-to-Speech Synthesis”, Published in: ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Date of Conference: 06-11 April 2025, DOI: 10.1109/ICASSP49660.2025.10890055, Date Added to IEEE Xplore: 07 March 2025. [6] Zhifang Guo, Yichong Leng, Yihan Wu, Sheng Zhao, Xu Tan, “Prompttts: Controllable Text-To-Speech With Text Descriptions”, Published in: ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Date of Conference: 04-10 June 2023, DOI: 10.1109/ICASSP49357.2023.10096285, Date Added to IEEE Xplore: 05 May 2023. [7] Miko?aj Babia?ski; Kamil Pokora; Raahil Shah; Rafa? Sienkiewicz; Daniel Korzekwa; Viacheslav Klimkov, “ On Granularity of Prosodic Representations in Expressive Text-to-Speech”, Published in: 2022 IEEE Spoken Language Technology Workshop (SLT), Date of Conference: 09-12 January 2023, DOI: 10.1109/SLT54892.2023.10022793, Date Added to IEEE Xplore: 27 January 2023. [8] Yan Shi; Jin Shi, Minchuan Chen, Chenfeng Miao, Ming Fang, Ning Cheng, “LEF-TTS: Lightweight and Efficient End-to-End Text-to-Speech Synthesis With Multi-Stream Generator”, Published in: ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Date of Conference: 06-11 April 2025, DOI: 10.1109/ICASSP49660.2025.10888129, Date Added to IEEE Xplore: 07 March 2025. [9] V. Madhusudhana Reddy, T. Vaishnavi, K. Pavan Kumar, “ Speech-to-Text and Text-to-Speech Recognition Using Deep Learning”, Published in: 2023 2nd International Conference on Edge Computing and Applications (ICECAA), Date of Conference: 19-21 July 2023, DOI: 10.1109/ICECAA58104.2023.10212222, Date Added to IEEE Xplore: 16 August 2023. [10] Ryo Nagakubo, Haruki Yamashita, Ryoichi Takashima, Misuzu Yasui, Tetsuya Takiguchi, “Training of VITS Model Reflecting the Duration of a Physically Unimpaired Speaker for a Text-to-speech System for a Person with a Stutter”, Published in: 2024 IEEE 13th Global Conference on zConsumer Electronics (GCCE), Date of Conference: 29 October 2024 - 01 November 2024, DOI: 10.1109/GCCE62371.2024.10760396, Date Added to IEEE Xplore: 28 November 2024. [11] Wooseok Han, Minki Kang, Changhun Kim, Eunho Yang, “Stable-TTS: Stable Speaker-Adaptive Text-to-Speech Synthesis via Prosody Prompting”, Published in: ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Date of Conference: 06-11 April 2025,DOI: 10.1109/ICASSP49660.2025.10890553,Date Added to IEEE Xplore: 07 March 2025. [12] Xinfa Zhu, Yuanjun Lv, Yi Lei, Tao Li, Wendi He, Hongbin Zhou, Heng Lu, Lei Xie,” Vec-Tok Speech: Speech Vectorization and Tokenization for Neural Speech Generation”, Published in: IEEE Transactions on Audio, Speech and Language Processing ( Volume: 33),Page(s): 1243 – 1254,DOI: 10.1109/TASLPRO.2025.3546559,Date of Publication: 27 February 2025. [13] Yi Ren, Chenxu Hu, Xu Tan, Tao Qin, Sheng Zhao, Zhou Zhao, Tie-Yan Liu,“FastSpeech 2: Fast and High-Quality End-to-End Text to Speech,”Published in: ICASSP 2021 - IEEE International Conference on Acoustics, Speech and Signal Processing,DOI: 10.1109/ICASSP39728.2021.9413889. [14] Jonathan Shen, Ruoming Pang, Ron J. Weiss, et al.,“Natural TTS Synthesis by Conditioning WaveNet on Mel Spectrogram Predictions,”Published in: ICASSP 2018 - IEEE International Conference on Acoustics, Speech and Signal Processing, DOI: 10.1109/ICASSP.2018.8461368.

Copyright

Copyright © 2026 Akula Sunitha, Korumilli Sri Lakshmi Asritha, Gudla Yamini Satya Durga, Chelluboina Bhavana, Vathadi Bhulakshmi, Isukapatla Papa. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.