Using CTCTextEncoder as the the EncoderDecoderASR tokenizer

[ai-fwd]

In a separate issue it was shown that we can use CTCTextEncoder as the tokenizer by copying the recipe here: https://huggingface.co/speechbrain/asr-wav2vec2-librispeech/blob/main/hyperparams.yaml

However, my decoder is a S2SRNNBeamSearcher and when using EncoderDecoderASR it tries to call self.tokenizer.decode_ids(token_seq) which CTCTextEncoder doesn't implement.

Am I doing something wrong or is this a known limitation? For now, i've updated the EncoderDecoderASR to check if it's a CTCTextEncoder and to call decode_ndim instead.

Thanks,