Validation Report

The validation report summarizes all validation results, gives recommendations for fixing errors and/or improving the overall quality of the speech corpus and gives an executive summary.

As a rough structure the validation report should contain:

• An executive summary listing the most prominent results of the validation.
• A short introduction to the speech corpus stating who produced it when and for what purpose.
• The results of the validation of the corpus documentation.
• All results of the automatic validation steps as listed in the validation contract, together with the methodology by which the results were achieved. If you list figures (percentage of errors), also give the appropriate confidence intervals^7.1. Long listings of erroneous files should be put into the appendix.
• All results of the manual validation together with a description of the techniques, the selection scheme^7.2, the statistics used and a profile of the participating validators. Again, don't forget to give confidence intervals for the results obtained.
• A list of the tools and programs used.
• Other relevant observations outside the required validation steps.
• Comments on how the quality of the corpus may be improved and what could be done better in future corpus productions.
• In the appendix: reference or corpus specification or validation contract on which this validation report is based on, listings of errors.

Try to prioritize the results and distinguish clearly between errors that can be and errors that cannot be repaired.

If you are performing regular validations on several releases of the same speech corpus, it might be a good idea to include a table summarizing the results of the actual and previous validations.

As an example you will find the validation report on the speech corpus WebCommand in appendix E.