Once the model is trained, you can perform the following command to execute the test steps using the test dataset:
(SpeechRecog)$python deepSpeech_test.py --eval_data 'test' --checkpoint_dir ./logs/
We evaluate its performance by testing it on previously unseen utterances from a test set. The model generates sequences of probability vectors as outputs, so we need to build a decoder to transform the model's output into word sequences. Despite being trained on character sequences, DS2 models are still able to learn an implicit language model and are already quite adept at spelling out words phonetically, as shown in the following table. The model's spelling performance is typically measured using CERs calculated using the Levenshtein distance (https://en.wikipedia.org/wiki/Levenshtein_distance) at the character level:
| Ground... |
|---|
