The examples assume a character set of size 28: a space, an apostrophe and 26 lower case letters. Transcripts will be normalized on the fly during training, as set in the YAML config templates, normalize_transcripts: lowercase. See Changing the character set for how to configure the character set and normalization. During validation, the predictions and reference transcripts will be standardized.

Training on multiple datasets can negatively affect training, only because the same word is transcribed differently across the datasets. The YAML config contains the entry standardize_text: true|false that allows to standardize the training transcripts on the fly. This happens only once, during data preparation phase, to avoid unnecessary slowdown between training steps. In principle, it is the same as WER Standardization. See WER calculation docs for more info on text standardization procedure.