The Librispeech directory structure is fixed, so a recursive transcript search seems excessive.
Additionally, in case of consecutive runs of ingest_librispeech, the recursive search lead to treating the output transcripts as inputs, if they were stored in a dataset subdirectory.
The Librispeech directory structure is fixed, so a recursive transcript search seems excessive. Additionally, in case of consecutive runs of ingest_librispeech, the recursive search lead to treating the output transcripts as inputs, if they were stored in a dataset subdirectory.