Speech Detection & Recognition Thresholds (SDT/SRT) & Word Recognition (WRS)

Speech Detection, Speech Recognition & Word Recognition Diagnostics

Speech audiometry constitutes an indispensable pillar of the comprehensive audiological evaluation. While pure-tone audiometry defines auditory sensitivity across discrete frequencies, speech stimuli evaluate the auditory system's capacity to detect, recognize, and resolve complex acoustic signals that carry human linguistic communication. A rigorous, doctoral-level understanding of speech audiometric protocols, psychometric validation, transducer acoustics, and masking equations is essential for identifying non-organic hearing loss, estimating hearing aid candidacy, and pinpointing retrocochlear or central auditory nervous system dysfunction.

---

Speech Detection Threshold (SDT) / Speech Awareness Threshold (SAT)

The Speech Detection Threshold (SDT)—synonymously termed the Speech Awareness Threshold (SAT)—is defined as the lowest intensity level, expressed in $\text{dB HL}$, at which a listener can just detect the physical presence of a speech signal $50\%$ of the time. Unlike speech recognition procedures, the SDT requires zero linguistic decoding, lexical identification, or repetition of the stimulus; the patient merely indicates awareness that vocal acoustic energy is present.

Clinical Stimuli & Presentation Methodology

• Continuous Discourse (Cold Running Speech): Rapid, uninflected narration delivered with uniform acoustic energy (peaking at $0\,\text{VU}$ on the volume unit meter) is the preferred stimulus. Connected discourse provides a continuous, redundant acoustic signal that facilitates rapid threshold convergence.
• Pediatric & Non-Verbal Tokens: For infants, toddlers, or non-verbal adult populations, familiar vocalizations—such as the patient's own name, repeated syllables (e.g., "ba-ba-ba"), or melodic vocal play—are effectively employed.
• Response Modalities: Standard behavioral responses include hand raising, button pressing, head turning, cessation of activity, or widening of the eyes (in pediatric behavioral observation).

Audiometric Cross-Check & Correlation with Pure-Tone Thresholds

The primary clinical purpose of the SDT is to serve as an independent biological validation of the pure-tone audiogram. Speech is a broadband signal, and speech detection is dictated by the frequency region of greatest hearing sensitivity within the core speech spectrum.

• The $\pm 5\,\text{dB}$ Rule: The SDT must agree within $\pm 5\,\text{dB}$ of the best (lowest) pure-tone threshold obtained between $250\,\text{Hz}$ and $1000\,\text{Hz}$ (most commonly $250\,\text{Hz}$ or $500\,\text{Hz}$).
• Clinical Example: If a patient presents with a precipitous high-frequency sensorineural hearing loss demonstrating thresholds of $20\,\text{dB HL}$ at $250\,\text{Hz}$, $25\,\text{dB HL}$ at $500\,\text{Hz}$, $60\,\text{dB HL}$ at $1000\,\text{Hz}$, and $85\,\text{dB HL}$ at $2000-4000\,\text{Hz}$, the expected SDT is between $15\,\text{dB HL}$ and $25\,\text{dB HL}$ (tracking the $20\,\text{dB HL}$ threshold at $250\,\text{Hz}$).
• Primary Indications: The SDT is indicated when a patient cannot perform speech recognition tasks due to chronological age (infants $< 12-18\,\text{months}$), developmental delay, profound intellectual disability, severe expressive/receptive aphasia, cognitive impairment, or severe language barriers.

---

Speech Recognition Threshold (SRT)

The Speech Recognition Threshold (SRT)—historically termed the Speech Reception Threshold—is defined as the lowest intensity level in $\text{dB HL}$ at which an individual can correctly repeat or identify $50\%$ of spondaic words from a closed or open set.

Spondee Words & Acoustic Characteristics

• Definition of a Spondee: A two-syllable word pronounced with equal vocal effort, stress, and duration on both syllables (spondaic meter: / -- -- /). Examples include: baseball, hotdog, airplane, cowboy, ice cream, cupcake, railroad, toothbrush, sidewalk, sunset, mousetrap, northwest.
• Psychometric Function Steepness: Spondees possess an exceptionally steep psychometric performance-intensity function, rising at approximately $10\%$ to $12\%$ per decibel near threshold. Because of this steep slope, the transition from $0\%$ to $100\%$ recognition occurs within a narrow $10-12\,\text{dB}$ window, rendering the $50\%$ threshold remarkably reliable and repeatable within $\pm 5\,\text{dB}$.
• Carrier Phrase Rules: A carrier phrase ("Say the word...") is optional for SRT testing. When utilized, the carrier phrase must be peaked at $0\,\text{VU}$ on the audiometer's volume unit meter, and both syllables of the spondee must independently peak at $0\,\text{VU}$. However, modern clinical guidelines frequently omit the carrier phrase during SRT determination to accelerate testing without sacrificing test-retest reliability.

Spondee Familiarization Protocol

Familiarization is an absolute, non-negotiable prerequisite of standard SRT testing (ASHA, 1988). The clinician presents a list of spondees suprathreshold (typically at $+30$ to $+40\,\text{dB SL}$ above estimated threshold or at a comfortably loud listening level) through the test transducer.

1. Aims of Familiarization:
   • Confirms that the target words exist in the patient's active vocabulary.
   • Ensures the patient can accurately execute the motor speech production required to repeat each word intelligibly.
   • Clarifies the patient's idiosyncratic pronunciation patterns so the clinician can score responses accurately.
   • Reduces vocabulary uncertainty, shifting the task from an open-set lexical search to a closed-set recognition task, thereby sharpening psychometric slope.
2. Exclusion Criteria: Any word that the patient fails to recognize, hesitates over, or cannot repeat correctly during suprathreshold familiarization is permanently removed from the test set for that patient.

Threshold Search Procedures: Hughson-Westlake vs. ASHA (1988)

Two primary protocols are utilized to determine the SRT:

• Modified Hughson-Westlake (Ascending) Method: Adapted from pure-tone audiometry. The clinician introduces a spondee below estimated threshold and ascends in $5\,\text{dB}$ steps until a correct response is given, then descends $10\,\text{dB}$ ("down 10, up 5"). The threshold is the lowest dial setting at which the patient correctly repeats at least 3 out of 6 words (or $50\%$ of presentations).
• ASHA (1988) Descending Protocol: A standardized statistical procedure designed to eliminate clinician bias. Testing begins at a suprathreshold level where the patient correctly repeats several words. The level is decreased in fixed steps ($2\,\text{dB}$ or $5\,\text{dB}$ decrements), presenting a fixed number of spondees per level down to the point where the patient misses all words. The threshold is calculated via a mathematical formula: $\text{SRT} = \text{Starting Level} - \text{Total Correct Words} + \text{Correction Factor}$ For the $5\,\text{dB}$-step version (5 words per level), the ASHA formula is the starting level minus the total words correct plus a $2\,\text{dB}$ correction factor.

---

The SRT-PTA Agreement Check & Clinical Discrepancy Analysis

The most fundamental quality-control metric in clinical audiometry is the comparison between the SRT and the Pure-Tone Average (PTA). Because speech acoustics encompass energy across the mid-frequency spectrum, speech thresholds must biologically track pure-tone thresholds.

Standard 3-Frequency PTA vs. Fletcher Two-Frequency Average

• Standard 3-Frequency PTA: The arithmetic mean of air-conduction thresholds at $500\,\text{Hz}$, $1000\,\text{Hz}$, and $2000\,\text{Hz}$: $\text{PTA}_3 = \frac{\text{AC}_{500} + \text{AC}_{1000} + \text{AC}_{2000}}{3}$
• Acceptable Agreement: In flat or gently sloping configurations, the SRT must agree with the 3-frequency PTA within $\pm 6\,\text{dB}$ (clinically acceptable within $\pm 8-10\,\text{dB}$).
• Fletcher Two-Frequency Average: When pure-tone thresholds slope steeply (defined as an inter-octave threshold disparity $\ge 20\,\text{dB}$ between adjacent speech frequencies), the 3-frequency PTA artificially overestimates the degree of speech handicap. Under these circumstances, the audiologist must calculate the Fletcher average—the arithmetic mean of the best two thresholds among $500$, $1000$, and $2000\,\text{Hz}$ (most often $500\,\text{Hz}$ and $1000\,\text{Hz}$): $\text{PTA}_{\text{Fletcher}} = \frac{\text{Best AC Frequency} + \text{Second Best AC Frequency}}{2}$ The SRT will consistently agree within $\pm 6\,\text{dB}$ of the Fletcher average rather than the 3-frequency PTA.

Clinical Diagnostic Discrepancies

When the SRT and PTA demonstrate a discrepancy exceeding $10-12\,\text{dB}$, the discrepancy represents a major diagnostic red flag requiring immediate investigation.

---

Word Recognition Testing (WRS / SDS)

Word Recognition Score (WRS)—historically termed Speech Discrimination Score (SDS)—is a suprathreshold, supraliminal test that quantifies the percentage of phonetically or phonemically balanced monosyllabic words correctly identified and repeated at a fixed, comfortable, or optimal presentation level.

Standardized Monosyllabic Word Lists

Monosyllabic test lists are meticulously constructed so that the distribution of phonemes within each list matches the relative frequency of phonemes occurring in conversational English.

• CID W-22 (Central Institute for the Deaf): Developed by Hirsh et al. Comprises four 50-word lists. Words are phonetically balanced and represent highly familiar, everyday adult vocabulary. Because the vocabulary is simple, CID W-22 functions are slightly less difficult and exhibit slightly shallower slopes than NU-6.
• NU-6 (Northwestern University Auditory Test No. 6): Developed by Tillman and Carhart. Comprises four 50-word lists arranged in four randomizations. Lists are constructed around consonant-nucleus-consonant (CNC) monosyllabic words built from the consonant-nucleus-consonant word lists of Lehiste and Peterson. NU-6 is the contemporary standard for adult audiological assessment and retrocochlear rollover diagnostics.
• PBK-50 (Phonetically Balanced Kindergarten): Formulated by Haskins. Comprises three 50-word lists with linguistic and lexical items selected from kindergarten-level spoken vocabularies. Designed for pediatric patients (developmental age $5-7\,\text{years}$) or adult patients with significant intellectual disabilities or restricted vocabularies.

Presentation Modality: Recorded Speech vs. Monitored Live Voice (MLV)

Carrier Phrase & Presentation Level Optimization

• Carrier Phrase Requirement: For monosyllabic word recognition testing, the carrier phrase "Say the word..." must precede every test item. The carrier phrase serves two vital functions: (1) it alerts the patient to prepare for the test token, and (2) it provides an acoustic calibration anchor, allowing the clinician (under MLV) to monitor and peak their voice at $0\,\text{VU}$, permitting the subsequent test word to fall naturally with conversational stress.
• Presentation Level Strategies:
  • Standard Suprathreshold Level: Routinely administered at $+30\,\text{dB SL}$ to $+40\,\text{dB SL}$ relative to the patient's SRT, ensuring the acoustic tokens are delivered with sufficient audibility to overcome threshold loss.
  • Conversational Speech Level: Administered at fixed levels of $60\,\text{dB HL}$ to $65\,\text{dB HL}$ to evaluate communication performance under typical real-world listening dynamics.
  • PB-Max Search: Administered at high levels (e.g., $80-90\,\text{dB HL}$ or $\text{UCL} - 5\,\text{dB}$) to determine the patient's maximum speech resolution ceiling for amplification fitting and rollover evaluation.

---

Binomial Distribution & The Thornton-Raffin Critical Difference Tables

A critical competency for the clinical audiologist is determining whether a percentage change in word recognition represents a true biological shift (e.g., progressive retrocochlear lesion, ototoxic decline, or true hearing aid benefit) or merely random statistical sampling error.

The Binomial Nature of Speech Recognition

Word recognition testing consists of $n$ independent, binary trials (the patient repeats the word either correctly or incorrectly). Consequently, test scores follow a binomial distribution, not a continuous Gaussian distribution. The standard error ($\sigma$) of a proportion $p$ based on $n$ test words is expressed mathematically as: $\sigma = \sqrt{\frac{p(1 - p)}{n}}$ Because variance is inversely proportional to list length ($n$):

• 50-Word Lists: Provide narrower confidence intervals and substantially higher statistical power.
• 25-Word (Half) Lists: Exhibit dramatically wider confidence intervals. A 25-word list doubles binomial variance, increasing the probability of false-positive and false-negative clinical decisions.

Thornton & Raffin (1978) Critical Difference Model

Thornton and Raffin established $95\%$ critical difference confidence intervals defining the exact range within which a second test score must fall to be considered statistically equivalent to the first score ($p > 0.05$).

Scores on a 25-word list move in 4% steps, so the table uses initial scores that are possible on both list lengths (values from Thornton and Raffin's 1978 tables).

Clinical Interpretation Rule: If a patient obtains an initial score of $80\%$ on a 25-word list, a follow-up score must drop below $56\%$ or rise to $100\%$ before an audiologist can conclude with $95\%$ statistical confidence that a true change in auditory capability has occurred. An apparent drop from $80\%$ to $64\%$ on a 25-word list is entirely attributable to binomial sampling error.

---

Clinical Masking for Speech Audiometry

Whenever high-intensity speech energy is delivered to a test ear, the acoustic signal can cross the cranial vault via bone conduction and stimulate the contralateral, non-test cochlea. Contralateral masking is required to prevent cross-hearing from creating an invalid, shadow speech result.

Interaural Attenuation (IA) for Speech Stimuli

• Supra-Aural Headphones (TDH-39 / TDH-49): $\text{IA} = 40\,\text{dB}$.
• Insert Earphones (ER-3A / ER-5A): $\text{IA} = 60\,\text{dB}$.

Speech Masking Rules

1. Masking Rule for SRT: Contralateral masking is mandatory whenever the SRT of the test ear (TE) minus interaural attenuation equals or exceeds the best (lowest) bone-conduction threshold of the non-test ear (NTE) across the speech range ($500$, $1000$, or $2000\,\text{Hz}$): $\text{SRT}_{\text{TE}} - \text{IA} \ge \text{Best } \text{BC}_{\text{NTE}}$
   • Initial Masker Level ($M_{\text{initial}}$): $M_{\text{initial}} = \text{SRT}_{\text{TE}} - \text{IA} + \text{Largest } \text{ABG}_{\text{NTE}} + 10\,\text{dB Safety Margin}$ Alternatively, presenting Speech Noise at $\text{SRT}_{\text{NTE}} + 10\,\text{dB}$ provides effective initial masking when the non-test ear has no conductive air-bone gap.
2. Masking Rule for Word Recognition Scores (WRS): Because word recognition is administered at suprathreshold levels ($+30$ to $+40\,\text{dB SL}$), the risk of cross-hearing is substantially elevated. Masking is mandatory whenever the speech presentation level ($PL_{\text{TE}}$) minus interaural attenuation equals or exceeds the best bone-conduction threshold of the non-test ear in the speech frequency range: $PL_{\text{TE}} - \text{IA} \ge \text{Best } \text{BC}_{\text{NTE}}$

Masking Stimulus: Speech Noise vs. White Noise

• Speech-Shaped Noise (Speech Noise): The internationally standardized masking stimulus for speech audiometry (ANSI S3.6). Speech Noise is a broadband noise filtered to match the Long-Term Average Speech Spectrum (LTASS). It exhibits a uniform, flat energy distribution from $100\,\text{Hz}$ to $1000\,\text{Hz}$, followed by a steep $-12\,\text{dB}$ per octave roll-off at frequencies above $1000\,\text{Hz}$.
• Acoustic Superiority: Speech Noise concentrates its acoustic masking power precisely in the spectral bandwidth where speech energy is greatest. Using White Noise (equal energy per Hertz) or Pink Noise ($-3\,\text{dB/octave}$) is acoustically inefficient, requiring excessive overall sound pressure levels that cause premature overmasking and patient discomfort.