Towards Improved Speech Recognition through Optimized Synthetic Data Generation

Yanis Perrin, Gilles Boulianne

Supervised training of speech recognition models requires access to transcribed audio data, which often is not possible due to confidentiality issues. Our approach to this problem…