Should we return WER per vendor in addition to WER per transcript (#32)? Consider: - The vendor information may not persist in the transcript - How to handle multiple hypothesis per vendor - do we simply average the WER?
