CommonVoice

The sequence-to-sequence approach is widely used in speech recognition (SR) nowadays, and many research works are dedicated to show that their capabilities relying on a single architecture often match or are even better than traditional hybrid or CTC systems with separately optimized components.