Do the new CTC decoders (CTCBaseSearcher, etc.) assume vocab_list has to be a list of single-character phones?

[tshmak]

I noticed that when i used both CTCPrefixBeamSearcher and TorchAudioCTCPrefixBeamSearcher, the output CTCHypothesis.text have all the phones joined together without space (c.f. this line). Therefore if I use something like the two-letter ARPABET, the spaces delineating the arpabets are all gone. Is this deliberate? Is it possible to support multi-lettered phones in the future?