Word recognition rate
The word recognition rate is a measure for assessing the accuracy of a speech recognition system. The word chain recognized by the system is compared with the actually spoken chain and the number of recognized words is determined.
In addition to the recognition speed, which is specified as a real-time factor (EZF), and the word accuracy , the word recognition rate is the essential measure for assessing the quality of the speech recognition system.
It is defined as
where N is the number of words in the reference and C is the number of correctly recognized words. In contrast to word accuracy, the word recognition rate cannot go negative.
A small example shows the calculation:
| Spoken sentence | Once | argued | themselves | North wind | and | Sun | |
|---|---|---|---|---|---|---|---|
| Recognized words | First | argued | North wind | themselves | and | Sun | |
| Type of error | S. | D. | I. |
The result is a recognition rate of 66.6%, since the insertions are not regarded as errors.
In the evaluation of speech recognizers, word accuracy is more widespread than the recognition rate, since the word recognition rate can be optimized as required by outputting as many words as possible at each point in the recognized chain. In contrast, the degree is widely used in medical speech processing to investigate the intelligibility of speech and voice disorders .