| _version_ | 1866901451836489728 |
|---|---|
| author | Gitanjali Salunke Gaytri Panchal Sujata Sunkewar |
| author_facet | Gitanjali Salunke Gaytri Panchal Sujata Sunkewar |
| contents | Podcasts have emerged as a widely adopted medium for information dissemination across domains such as technology, education, health, and entertainment. However, their extended duration and unstructured conversational format make efficient content discovery challenging. This paper presents an automated Podcast Summarization and Topic Classification system that integrates Automatic Speech Recognition (ASR) and transformer-based Natural Language Processing (NLP) models to convert raw audio into structured textual insights. The proposed framework employs Whisper ASR for accurate speech-to-text transcription, followed by preprocessing techniques including tokenization, stop-word removal, and normalization. Abstractive summarization is performed using the BART transformer model to generate concise and semantically coherent summaries. Topic classification is achieved using supervised machine learning algorithms, including Logistic Regression, Decision Tree, Random Forest, and contextual modeling BERT. Experimental evaluation demonstrates reliable transcription accuracy and improved classification performance across multiple podcast genres. The system significantly reduces listening time, enhances accessibility, and improves content discoverability. The modular architecture ensures scalability and provides a practical solution for intelligent audio content analysis in academic and real-world applications. |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_20124605 |
| institution | Zenodo |
| language | |
| publishDate | 2026 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | Automatic Podcast Summarization and Topic Classification Using ASR and Transformer-Based Models Gitanjali Salunke Gaytri Panchal Sujata Sunkewar Podcast Summarization Automatic Speech Recognition Transformer Models Topic Classification Natural Language Processing Machine Learning. Podcasts have emerged as a widely adopted medium for information dissemination across domains such as technology, education, health, and entertainment. However, their extended duration and unstructured conversational format make efficient content discovery challenging. This paper presents an automated Podcast Summarization and Topic Classification system that integrates Automatic Speech Recognition (ASR) and transformer-based Natural Language Processing (NLP) models to convert raw audio into structured textual insights. The proposed framework employs Whisper ASR for accurate speech-to-text transcription, followed by preprocessing techniques including tokenization, stop-word removal, and normalization. Abstractive summarization is performed using the BART transformer model to generate concise and semantically coherent summaries. Topic classification is achieved using supervised machine learning algorithms, including Logistic Regression, Decision Tree, Random Forest, and contextual modeling BERT. Experimental evaluation demonstrates reliable transcription accuracy and improved classification performance across multiple podcast genres. The system significantly reduces listening time, enhances accessibility, and improves content discoverability. The modular architecture ensures scalability and provides a practical solution for intelligent audio content analysis in academic and real-world applications. |
| title | Automatic Podcast Summarization and Topic Classification Using ASR and Transformer-Based Models |
| topic | Podcast Summarization Automatic Speech Recognition Transformer Models Topic Classification Natural Language Processing Machine Learning. |
| url | https://doi.org/10.5281/zenodo.20124605 |