R17 Normalise reference transcript into internal extensible format

[EyalLavi]

We want clients to be able to benchmark using different criteria: text only, text and punctuation, timings, speaker identification. The tool needs to identify the elements in the reference transcript for benchmarking. We also want to avoid forcing users to make complicated reference files. So we need a way to convert different formats into an internal format. The internal format could be a list of tokens (words) with metadata (TBC).
Proposed input formats:

• Plain text
• CTM
• MLF