Publications/Speech Processing, Recognition & Audio Security
Journal · 2025
Conversations in the Wild: Data Collection, Automatic Generation and Evaluation
Nimra Zaheer, Agha Ali Raza, Mudassir Shabbir
Computer Speech & Language, vol. 89, 2025 · IF 3.4
Summary
Systems that work out who spoke when need large collections of annotated conversations, which exist for only a few languages such as English, Mandarin, and French. We propose two language-independent algorithms that assemble realistic multi-speaker conversations from ordinary, non-conversational speech recordings, and release 283 hours of such synthetic conversations from 50 speakers. We also compile the first spontaneous conversational dataset for Urdu: 24 hours and 212 speakers drawn from public talk shows. On speaker diarization we report the first Urdu baselines, with error rates of 25% for models trained on the synthetic conversations and 29% for natural ones.
Cite this paper
Nimra Zaheer, Agha Ali Raza, Mudassir Shabbir. Conversations in the Wild: Data Collection, Automatic Generation and Evaluation. Computer Speech & Language, vol. 89, 2025.
@article{zaheer2025conversations,
author = {Nimra Zaheer and Agha Ali Raza and Mudassir Shabbir},
title = {{Conversations in the Wild: Data Collection, Automatic Generation and Evaluation}},
journal = {Computer Speech \& Language},
volume = {89},
year = {2025},
url = {https://aghaaliraza.com/publications/2024--CSL--Conversations_in_the_Wild_Data_Collection_Automatic_Generation_and_Evaluation.pdf}
}
- Dissecting ASR Failures in Low-Resource South Asian LanguagesInterspeech 2026, Sydney, Australia
- WER We Stand: Benchmarking Urdu ASR ModelsCOLING 2025, Abu Dhabi, UAE
- Deepfake Defense: Constructing and Evaluating a Specialized Urdu Deepfake Audio DatasetFindings of ACL 2024, Bangkok, Thailand
- Speech Emotion Recognition for the Urdu LanguageLanguage Resources and Evaluation (LRE), 2023
- Using Self-Attention DNNs to Discover Phonemic Features for Audio Deep Fake DetectionASRU 2021, Cartagena, Colombia