After a successful model test you can find a detailed report of the recognition accuracy here.
To best reflect the recognition performance, Sam uses multiple, ranked results for the tests.
A correctly recognized word or sentence will be scored with the confidence score the word achieved. If a word is recognized correctly but another, wrong, result is ranked higher, the accuracy of that recognition will be 0%.
The displayed overall recognition rate is the average of all confidence rates. The shown overall accuracy rating shows the average over all accuracy scores which represents the likelihood that given any word in any sentence, this word has been recognized correctly.
Sam will list recognition accuracy for each word individually.
If you have samples containing more than one word they will be segmented during recognition. Each word will be scored individually (although the different words of course still influence each other).
This section lists each prompt as "sentence".
Prompts that were recorded more than once are combined.
In the files section you can see the recognition results for each file. Each file will list the 10 most likely results in the details pane when you select it.
When you identify problematic samples, you can re-record (or remove) them by selecting them and clicking on .
You can sort the files by each column simply by clicking on the column header. This way it is very easy to find bad samples by sorting by recognition rate.