Mendel (2008) notes "speech perception" is not directly measurable. This issue poses a greater problem when assessing children, more so, than adults. Mendel reminds us speech perception is an abstract theoretical construct more so than an objective, concrete metric.
Indeed, the most common speech tests used in the clinic measure monosyllabic word recognition ability, or perhaps phoneme or sentence recognition. Thus, a battery approach is warranted and is typically used (combining word recognition scores with speech reception thresholds and perhaps sentences...) to create an overall speech perception approximation. Mendel notes many variables impact pediatric speech perception results, including hearing loss, vocabulary, language competence, chronological age, alertness, attention, mood, cognitive abilities, motivation, fatigue, and other factors such as the type of task involved. Luckily, as children age, the difference between competence and performance decreases. Test administration may present other sources of variation and error due to stimulus familiarity, phonetic and phonemic balancing, carrier phrase (or not), presentation format, presentation loudness level, length list, presentation in quiet and/or noise, recorded versus live, open versus closed-set, etc.
Newer tests of speech perception address these concerns with better validity and reliability than the classic, well-known tests. Mendel reminds us that validity is not an absolute, but rather, is more a matter of degree. The four types of validity to be considered are (1) predictive, (2) content, (3) construct, and (4) face. Reliability indicates how repeatable the results are. Importantly, tests can be reliable without being valid; however, to be valid, a test must be reliable.
Mendel reviews the Neighborhood Activation Model (NAM) and the concept that words exist in a "lexical neighborhood." Basically, for a particular word, all options (other words) that are "one phoneme away" (with respect to substitution, insertion, or deletion) represent the lexical neighborhood. Easy words have few lexical neighbors, hard words have lots of neighbors. As such, the Lexical Neighborhoods Test (LNT) and the Multisyllabic Lexical Neighborhoods Test (MLNT) are essentially "top-down" processing tests based on the lexical neighborhood theory. With respect to sentence tests, the Bamford-Kowal-Bench and the HINT-C are reviewed, as is the Listening in Spatialized Noise (LISN) test.
For More Information, References and Recommendations:
Cameron, S. and Dillon, H. (2007): The Listening in Spatialized Noise Sentence-Test (LISN-S): Test-retest Reliability Study. International journal of Audiology, 46, 145-153.
Cameron, S. Dillon, H and Newall, P. (2006): Development and Evaluation of the Listening in Spatialized Noise Test. Ear and Hearing. 27(1), 30-42.
Mendel LL., (2008) Current Considerations in Pediatric Speech Audiometry. International Journal of Audiology. Vol 47, pgs 546-553.
"Researcher developing new method for hearing loss assessment"
http://www.eurekalert.org/pub_releases/2007-09/pu-rdn_1090607.php