Short answer
From a clinical audiological perspective, we talk about impaired speech understanding/recognition/discrimination.
Background
From a psychophysical point of view, and specifically in the realm of clinical audiology where we deal with speech testing for people with impaired hearing, there are a number of speech tests available that measure slightly different things. These speech tests can be performed in silence where words may be misunderstood due to the presentation of background noise, or because of hearing impairments.
Speech understanding can be loosely defined as knowing the meaning of what was said by the target speaker. This involves listening, recognizing speech and interpreting it at a higher level using associative thinking. An example sentence of a simple listening test would be 'The students are protesting against nuclear weapons'. The sentence is meaningful and the relation between the elements of the sentence can help the listener to understand the sentence, even when parts of the sentence are not heard due to for instance background noise, or hearing loss. For instance, if you know it's about a protest of students and you hear 'nuclear', but the weapons is not heard, it can still be interpreted from the context.
However, more is up to this when you change the task from simply repeating the sentence, to asking :'Why do the students protest'? The answer should be that nuclear weapons are a danger for mankind (for instance), but if for some reason the test subject is unaware of that, or doesn't know what nuclear weapons are, there is impaired speech understanding.
If it is not the understanding being tested, but the listener simply has to repeat back what was heard, we talk about speech recognition testing. When a word in the sentence was simply not heard due to noise for instance, and can't be deduced from the sentence, say in a digit-in-noise test (Smits et al, 2013) where only random numbers are being used without context ('two -nine-eight) there is impaired speech recognition.
If a closed set of words is used to generate sentences that is beknown to the listener, e.g., the Matrix test (Houben et al., 2014), then it is not speech understanding or recognition being tested, but strictly spoken it is word discrimination being measured instead. Errors are now, well no wonder, called impaired speech discrimination.
References
- Smits et al., JASA (2013); 133: 1693
- Houben et al., Int J Audiol (2014); 53(10)