Speaker Profiles

The characteristic features of each speaker should be collected in the speaker profile set of the corpus. This may be either a file for each speaker (recommended) or a table summarizing the features per speaker in one line or column. To be machine-readable the speaker profiles have to be standardized (parsable), optimally in XML (see section , p. ).

If you are producing speech corpora on a regular basis, you might consider including speaker profiles of your corpora into a database system. Keep the mapping from speaker names to IDs separate from this database (see also chapter for a discussion of data protection).