These cookies ensure basic functionalities and security features of the website, anonymously. Necessary cookies are absolutely essential for the website to function properly. We compile the training data from a wide variety of sources in order to achieve a high level of generalization. TRAININGĪppTek trains all ASR models on very large collections of annotated audio data. The toolkit is built upon the TensorFlow backend and allows flexible and efficient specification, training, and deployment of different neural models. The models are trained using the RETURNN toolkit - a software package for neural sequence-to-sequence models, developed jointly by of the RWTH Aachen University, Germany and AppTek. OVERVIEWĪppTek’s acoustic models are backed by bi-directional recurrent neural networks with LSTM units. When we build ASR systems for academic tasks, following the comparable training and evaluation conditions as other ASR teams in the community, we achieve state of the art results on popular US English benchmark tasks like LibriSpeech (5.5% on “test-other”) or Switchboard (11.7% on “Hub5 2000 eval”). Model Technical Details PERFORMANCE METRICSĪppTek’s ASR models achieve approximately the same accuracy on real world data as the top cloud service providers.

Chinese speech to text software