Skip to content

Releases: GeorgeFedoseev/DeepSpeech

21% WER on voxforge-ru-test set of Russian speech

18 Jun 08:55
5b361ec
Compare
Choose a tag to compare

For inference

  • Ready acoustic model (TensorFlow frozen graph) output_graph.pb

To continue training from checkpoint

  • Checkpoint directory to continue training from checkpoint (note that n_hidden training parameter should be 2048)

Language model

  • Language Model (lm.binary, trie and alphabet.txt)