Faetar Low-Resource ASR Challenge 2025: A New Opportunity for Speech Recognition Research

Join the Faetar Low-Resource ASR Challenge 2025 to address common issues in archival collections of speech data and contribute to the preservation and learning of the endangered Faetar language. Register now to access the Faetar ASR Benchmark Corpus and compete in four tracks focused on ASR architectures, pre-trained models, unlabelled data, and dirty data.

The Faetar Low-Resource ASR Challenge 2025 is now open for contributions. This challenge focuses on addressing issues commonly found in archival collections of speech data such as noisy field recordings, lack of standard orthography, and limited transcribed data. The challenge uses the Faetar ASR Benchmark Corpus, which represents the majority of all archived speech recordings of Faetar in existence. Four tracks are available, including a constrained ASR track and three thematic tracks focusing on pre-trained models, unlabelled data, and dirty data. Submissions will be evaluated based on phone error rate (PER) on the test set. For more information and to register, visit the challenge website.

Tags: Faetar Low-Resource ASR Challenge 2025, speech recognition, archival collections, noisy field recordings, standard orthography, transcribed data, Faetar ASR Benchmark Corpus, constrained ASR, pre-trained models, unlabelled data, dirty data, phone error rate, Interspeech 2025