Noisy join mixed ASR
Problem with common ASR dataset, each audio sample is a monolanguage, but we can generate an audio sample with multilanguage using ASR and Force Alignment models. Notebooks at https://github.com/huseinzol05/malaya-speech/tree/master/data/noisy-join-mixed-asr
Figure like below,
Generated samples ~601 hours.