WorldSpeech: A Multilingual Speech Corpus from Around the World
Addresses the data scarcity bottleneck for low-resource languages in ASR, providing a large-scale resource to improve multilingual speech recognition.
WorldSpeech introduces a 65k-hour, 76-language speech corpus, achieving a 63.5% average relative WER reduction when fine-tuning ASR models on 11 diverse languages.
Automatic speech recognition (ASR) performs well for high-resource languages with abundant paired audio-transcript data, but its accuracy degrades sharply for most languages due to limited publicly available aligned data. To this end, we introduce WorldSpeech, a 24 kHz multilingual speech corpus comprising 65k hours of aligned audio-transcript data across 76 languages, collected from diverse public sources including parliamentary proceedings, international broadcasts, and public-domain audiobooks. For 37 languages, WorldSpeech provides more than 200 hours of aligned speech, with 28 exceeding 500 hours and 24 surpassing 1k hours. Fine-tuning existing ASR models on WorldSpeech results in an average relative Word-Error-Rate reduction of 63.5% across 11 typologically diverse languages.