Hybrid Transducer and Attention based Encoder-Decoder Modeling for Speech-to-Text Tasks