SRT vs WRS: speech reception and word recognition

The speech reception threshold, or SRT, and word recognition testing both use speech as a stimulus, but they answer different questions. The SRT asks how quiet speech can get before it stops being understandable, giving a threshold, in dB HL, that should track a person's overall degree of loss. Word recognition testing asks how clearly speech is understood once it is loud enough to be comfortably audible, giving a percentage score that reflects clarity of understanding rather than sensitivity.

The SRT is measured using spondees, two-syllable words with roughly equal stress on both syllables, such as baseball or hotdog, presented at descending levels until the patient can only correctly identify about half of them. The result is a threshold in dB HL, conceptually similar to a pure-tone threshold but obtained with speech instead of tones.

Because the SRT is a threshold measured with a broadband signal, it should agree reasonably closely with a pure-tone measure of degree of loss. For a flat audiogram, that comparison is the standard three-frequency pure-tone average of 500, 1000, and 2000 Hz, and agreement within about 5 dB is considered normal, with up to about 10 dB still commonly accepted. For a sloping or rising audiogram, the SRT is compared instead to a two-frequency average of the best two of those three thresholds, because a broadband speech signal is naturally driven by whichever frequencies the ear hears best, and expecting it to match a flat three-frequency average on a sloping loss sets up a mismatch that is not actually abnormal.

A disagreement greater than roughly 10 to 12 dB between the SRT and the appropriate pure-tone comparison is unusual enough to raise a flag, most often for a nonorganic, or functional, hearing loss, but sometimes for a testing error such as comparing the SRT to the wrong pure-tone average for that audiogram's shape. The single most common mistake here is expecting a sloping loss to produce an SRT that matches the flat three-frequency average; a better-than-expected SRT on a sloping audiogram is the normal result of that shape, not a sign of exaggeration.

Word recognition testing presents a list of single, one-syllable words, phonetically balanced, at a comfortable level well above the patient's SRT, typically 25 to 40 dB above it or at the patient's most comfortable listening level, and scores the percentage repeated correctly. Because a standard list has a fixed number of words, usually 25 or 50, only certain scores are actually obtainable: a 50-word list scores in 2 percent steps, so 89 percent is not a value that can ever be produced by that list.

Word recognition scores are described with a set of rough descriptors: roughly 90 to 100 percent is considered excellent, 78 to 88 percent good, 66 to 76 percent fair, 54 to 64 percent poor, and below about 52 percent very poor, though different sources draw the boundaries slightly differently, especially around the good-to-fair line. A word recognition score obtained only at one level does not by itself establish a person's maximum score, called PB max; a full search across multiple presentation levels is needed before word recognition scores can be used to look for rollover or to suggest a retrocochlear site of lesion.