Depending on the speaker ID task, IDOL Speech Server can return the following warnings.
Task | Warning message | Recommended action |
---|---|---|
SpkIdTrain , SpkIdTrainWav , SpkIdTrainStream |
Can only share base component variance if UBM is set | Set the base model to use, or switch off component sharing. |
Can only share base component means if UBM is set |
Set the base model to use, or switch off component sharing. |
|
Problem opening audio feature file | Check that all the feature files listed in the DataList file exist, and are non-empty.
|
|
Could not read feature file header – invalid format? | Ensure that all input feature files are in the correct format (that is, generated by using the SpkIdFeature task).
|
|
Feature frame inconsistency detected with file | Ensure that all input feature files being used to train the templates were trained with the same configuration (for example, that nFeatures was the same in all cases).
|
|
Input label file is longer than audio | Check that the label file being used to mark speaker sections matches the audio in terms of the total duration. | |
A smaller proportion of the audio was identified as speech than expected | Ensure that enough actual speech data is available for training. Check the audio file to validate whether a large proportion was indeed silence or noise. If not, you might need to increase the LowEnergyValue setting in the normalizer module when you generate features.
|
|
SpkIdDevelFinal
|
No positive examples for template |
Ensure that you provide positive audio examples (data from each speaker) for each speaker template. |
No positive examples for template) | Ensure that you provide negative audio examples (data from speakers other than just the trained speaker) for each speaker template. | |
No development scores found for speaker – threshold not set |
Ensure that all speaker templates are included when you run the development tasks ( |
|