Suprathreshold Speech Recognition II Flashcards Preview

HESP606 > Suprathreshold Speech Recognition II > Flashcards

Flashcards in Suprathreshold Speech Recognition II Deck (22):

Describe closed-message tests. What words are used? How many foils are there? How are the foils chosen? How is the guess rate determined?

  • General Description
    • Monosyllabic tests
    • small number of foils - alternative choices
    • foils chosen to be easily confusable with target
    • opportunity for guessing
      • 6 choices/item -> guess rate of 16.6%


Describe the Fairbanks Rhyme Test. What kind of test is it? What were their lists? How were the words related? What is the listener provided with? Phonetically balanced? What does it probe? How can it be administered?

  • Closed-message test
  • Fairbanks Rhyme Test (1958)
    • 5 lists drawn from 250 monosyllables
    • all rhyming words between lists
    • listener is provided with word stem
    •         List 1  List 2  List 3  List 4  List 5
    • _at    mat     bat     sat      cat     rat
    • _in      tin      sin     shin     bin    chin
    • not phonetically balanced
    • probes differentiation of initial consonant
    • can be administered to groups


Describe the modified rhyme test. What kind of test is it? What is its purpose? How is it structured? What are the results with normal hearing listeners?

  • Closed-message test
  • Modified Rhyme Test (House et al., 1963)
    • purpose: evaluate communication systems with untrained listeners
    • structure: 6 lists, 50 words/ea, 6 choices/item
      • 25 items: varied initial (I) consonant
      • 25 items: varied final (F) consonant
    • listener has all 6 choices available
    • results with normal-hearing listeners
      • Good interlist equivalence, no learning effects
      • Scores on MRT are > PAL PB-50’s by 25%
      • Shift to 50% ot 60% with white noise


Describe the California Consonant Test. What kind of test is it? What is its purpose? What was its structure?

  • Closed-message test
  • California Consonant Test (Owens & Schubert, 1977)
    • purpose: develop a test for clinical populations
      • Especially people with high frequency sensorineural loss 
        • People with noise notch do well on NU-6 so this was developed 
        • People with high frequency hearing loss make place errors
    • structure: 100-item test, 4 choices/item
      • 36 items varied I consonant
      • 64 items varied F consonant - harder
      • items loaded with fricatives and sibilants


Describe the evaluation of the CCT. What was obtained? What were the slopes and asymptotes? What else was compared (2)? What did they find for each? 

  • Schartz & Surr (1979)
    • Obtained articulation functions for normal and hrg imp L’s
      • Slope = 1.6%/dB (normals) and 1.3%/dB (hrg imp) - more gadual - harder
      • Performance asymptotes at 50 dB SL re SRT
    • Compared performance on NU6 and CCT (hrg imp only)
      • High scores on NU6 (M = 87%); small s.d.
      • Lower scores on CCT (M = 68%); higher s.d.
    • Compared performance on 2 50-word forms
      • r between 1st half and 2nd half of list (.71); high variability


Describe the study by Clopper, Pison, and Tierney on open vs. closed sets. What did they investigate? What stimuli were used? What were the results? What was the conclusion?

  • Clopper, Pisoni, & Tierney (2006)
    • Investigated task demands and lexical issues with open vs. closed sets
      • Presented MRT and PB-50’s, but divided words into lexically easy and lexically hard
      • Words recorded by 5 different talkers
      • Varied # of alternatives and difficulty of alternative responses
    • Results:
      • Closed set performance > open-set
      • Lexically easy words > lexically hard words
      • # alternative choices and difficulty of choices influences performance
    • Conc: lexical knowledge and task demands influence performance on open and closed-set speech recognition tasks
      • Frequency and neighborhood both influence performance.


Describe the CID everyday sentences. What kind of test is it? What is its purpose? What was the structure? How is it scored? How does it compare to W-22?

  • Sentence Test
  • CID Everyday Sentences (Silverman & Hirsh, 1955)
    • purpose: estimate everyday hearing disability
    • Structure
      • 100 sentences, everyday speech, familiar vocabulary
        • predictable sentences - "walking is my favorite exercise"
      • Sentences are 2-12 words in length
      • 10 lists of 10 sentences/ each
    • listener’s task: identify all words in the sentence
    • Scoring: based on keywords in sentence
    • comparison with W-22 performance:
      • Hrg imp L’s: performance on CID sentences > W-22’s by 28%


Describe the Synthetic Sentence Identification test. What kind of test is it? What is its purpose? What stimuli were developed? How are stimuli presented? How is this test helpful with diagnosis?

  • Sentence test
  • Synthetic Sentence Identification Test(SSI) (Speaks & Jerger, 1965)
    • Purpose: identify patients with N. VIII lesions
      • need to vary temporal structure of speech; thus, need sentences
        • N VIII lesions - difficulty with timing characteristics of speech
      • Avoided use of contextual cues by developing synthetic sentences
        • Sentence length but with no contextual cues
        • Synthetic in terms of word order
    • Structure
      • Used common vocabulary; followed rules of syntax
      • Developed 3rd order approximations to sentences
      • Ex: “Forward march said the boy a”
      • Developed lists of 10 sentences/each
    • Presentation
      • Closed set, listener identifies sentence heard
      • Presented in background of single talker
        • In quiet would be able to pick out correct sentence
      • Varied message-to-competition ratio (MCR) from -40 dB -> + 20 dB
      • Competition goes into ipsilateral ear or contralateral ear (ICM - ipsilateral competing message, CCM)
      • Listener gets feedback (correct vs. incorrect)
    • diagnostic usefulness
      • helps distinguish brainstem vs. cortical lesions
        • Right cortex lesion - poor performance when target is in left ear


Describe Dichotic Sentence Identification. How are stimuli presented? What does the subject select? What modes are there? How many sentences presented? How does SNHL effet ability to do DSI? What are normal scores? Which mode has higher scores? Who does more poorly and why? Describe normal performance. Describe performance with cortical lesion.

  • Dichotic Sentence Identification (DSI)
    • 6 SSI sentences, 1 presented to each ear simultaneously, at 50 dB SL
    • participant selects 2 sentences from a list
    • 2  modes: directed mode and free
      • Directed: tell me R then L
      • Free: order doesn't matter
    • 10 sentences presented/ear
    • Relatively resistant to the effects of sensorineural hearing loss until the degree of loss exceeds 50 dB HL.
    • Normal scores >80 %, with right ear advantage seen
    • Directed mode scores >free report
    • Older listeners do poorly on this test, and often associated with memory problems (Gates et al., 2008)
  • Normal performance (above)
  • Cortical lesion: poor performance in ear opposite the lesion on CCM BS lesion: poor performance on ICM – both ears.

A image thumb

Describe the CUNY NST. What kind of test is it? What is its purpose? What is it structure? How is it presented? What are the results? What is the test useful for?

  • Nonsense Syllable Test
  • CUNY NST(Resnick, Dubno, Hoffnung, Levitt, 1975)
    • Purpose: determine absolute recognition ability of individual listeners
    • Structure
      • 91 different items
      • All possible consonants paired with /a, i, u/
      • CV or VC format
      • ex:  ba ---> ba, da, ga, la, ra, wa, ja, dja
      • 11 subtests with 7-9 syllables/each
    • presentation and results
      • Closed response format
      • No learning effects, high inter-list equivalence
      • No possibility of word understanding
    • usefulness of test
      • Can determine consonant confusions of individual -> developing rehabilitative program, benefit of hearing aids
      • Can see if hearing aids are helping to hear the consonants


What is the purpose of testing in noise? What are the procedural variations of testing in noise? What are the advantages and disadvantages of each?

  • Purpose
    • Measure performance in realistic situations
    • Identify deficits the person may experience
  • Procedural variations
    • Fixed signal-to-noise ratio (SPIN, AzBio)
      • Advantages
        • easy to interpret % correct score
      • Disadvantages
        • time-consuming, only 1 S/N ratio doesn’t reflect everyday situations
    • Adaptive procedures - change SNR as you go (HINT)
      • Advantages:  quick
      • Disadvantages: difficult to interpret (SNR)
    • Method of Constants - drop SNR every couple of items (Q-SIN, WIN)
      • Advantages: quick
      • Disadvantages: difficult to interpret (SNR)


Describe the Speech Perception in Noise test. What kind of test is it? What is the purpose of the test? What is its structure? How many sentences per list? What level are the stimuli presented at? What are the standardization results (reliability, list equivalence, performance of listeners)?

  • Sentence test in noise - Fixed SNR
  • Speech Perception in Noise (SPIN)
    • Originally developed by Kalikow, Stevens, & Elliott (1977)
      • R-SPIN (Bilger et al., 1984)
    • Purpose: compare person’s recognition of acoustic information to additional use of contextual cues
    • Structure
      • Sentences have low context (LP items) or high context (HP items)
      • 50 sentences/list - half are LP and half are HP (target word is last word of the sentence)
      • Present at 50 dB SL re babble threshold, in 12-talker babble at  +8 dB S/B ratio (simulates average noise conditions)  
    • standardization results (Bilgeret al., 1984)
      • Reliability: acceptable (HP: .91, LP: .85)
      • list equivalence: 8 forms in R-SPIN are equivalent
      • performance of listeners
        • Hrg imp: 92% on HP SPIN, 44% on LP SPIN
        • Normals: 100% on HP SPIN,  70% or higher on LP-SPIN


Describe the adaptive prcedure. What is the procedure? What is the adaptive rule? How is S/B calculated?

  • adaptive procedure (Dirks, Morgan, & Dubno, 1982)
    • procedure: fixed signal level, vary noise level
      • Ex: start with noise at 30 dB below signal level
      • Present 3 words/step in 2-dB steps
      • If 2/3 words correct -> increase noise level
      • If 2/3 words incorrect -> decrease noise level
    • adaptive rule: continue until 6 reversals in direction  
    • calculation of S/B for 50% criterion performance: mean of the midpoints of the final 4 excursions


Describe the Hearing in Noise Test. What kind of test is it? What is its purpose? What are requirements for development?

  • Sentence test in noise, adaptive procedure
  • Hearing in Noise Test (HINT)
    • Nilsson, Soli, & Sullivan (1994)
    • Purpose: develop a test of speech recognition  that avoids floor and ceiling effects
      • requires testing in noise with adaptive procedure
    • requirements for development
      • Produce reliable results
      • Multiple lists, equated in difficulty
      • No learning effect with materials
      • Representative of language
      • Sentence-length materials satisfy requirements


What is the structure of the HINT? What is the procedure? What does it measure? What is the adaptive rule? What were the results for normal hearing listeners? What is the test useful for?

  • structure of test
    • BKB sentences (Bamford-Kowal-Bench, 1979)  
    • Sentences on a list were phonetically balanced
    • 25 lists with 10 sentences/list, 3 practice lists
    • Scoring for keywords
    • Recordings: male talker 
    • Presented in speech spectrum noise
    • procedure
      • Measuring S-SRT in quiet and in fixed noise level of 72 dBA under earphones (for SF testing: noise is 65 dBA).
      • Listener repeats sentence presented
      • Adaptive rule: If entire sentence correct, speech level attenuated by 2dB, if incorrect increase level by 2 dB
        • Noise stays constant
  • Results reported in article (normals)
    • Mean SRT in quiet = 23.91 dBA
    • Mean S/N ratio at threshold in noise: -2.92 dB (i.e., SRT = 69.08; noise fixed at 72 dBA)
      • SNR that is required for 50% correct
    • Lists were equivalent
    • Reliability was high
    • Very popular test; available in many languages
    • Used to assess “functional hearing”
      • Good for hearing critical professions (firefighters, military, Police, FBI)
        • Require normal HINT performance
      • Evaluate performance for noise in 2 conditions: 
        • spatially separated noise (speech at 0° and noise at 90°) 
        • co-located  (speech and noise at 0° azimuth)


Describe the SIN Test and Q-SIN. What kind of tests are they? What was the purpose of development on the SIN test? What is the structure of the test? What does the Q-SIN assess? How many lists are there and sentences per list? How are they presented? What is the presentation level (earphones, sound field)? How is it scored? How is SRT calculated? What is SNR loss and how is it calculated?

  • Sentence tests in noise, modified method of constants
  • SIN Test (Fikret-Pasa, 1993 at Etymotic Research) and Quick-SIN Test (Killion, et al., 2004)
    • (*Modified method of constants – change SNR during the test; not adaptive and not fixed SNR)
    • Purpose of development: evaluate effects of compression ratio in hearing aids on speech intelligibility and quality
      • needed a test in noise with multiple presentation levels
    • Structure of test
      • IEEE sentences, multiple lists (“The birch canoe slid on the smooth planks”)
      • Original SIN presented at two signal levels in noise with 4talker babble (multiple SNRs)
  • QuickSIN (Killion et al., 2004)
    • Assesses “SNR loss” in a 1-minute test
    • 12 equivalent lists, each list has 6 sentences, 5 keywords/sentence (+6 extra lists)
    • Sentences on a list are presented in noise at 6 SNRs: 25, 20, 15, 10, 5, 0 (1 sentence/SNR)
    • Presentation level:
      • Under earphones: speech at 70 dB HL, start noise at SNR = +25 dB, increase noise level in 5 dB steps after each sentence
      • In Sound Field: speech at 50 dB HL; start noise at SNR = +25 dB,
    • Scoring -based on keywords in each sentence; extrapolate SNR for 50% correct
      • Uses Tillman-Olsen formula for calculating SRT
      • Starting SNR - # correct + 2.5 dB (correction factor)
      • Raw results: SNR required for 50% correct
      • Standard deviation for SNR estimate with one list is 1.4 dB  
    • Convert SNR for 50% correct to “SNR loss” - how much poorer is person compared to normal 
      • Individual SNR - Normal SNR (2 dB); short-cut: 25.5 - # correct
      • Compared to normal score to indicate degree of severity
        • SNR loss 0-2 dB = normal to near normal
        • SNR loss 2-7dB = mild
        • SNR loss 7-15 dB = moderate
        • SNR loss > 15 dB = severe


Describe the Words in Noise Test. What kind of test is it? What stimuli does it use? What is the presentation level? How many words per level? When do you stop presentation? How is SNR 50% calculated? What is the performance for normal listeners? Hearing impaired? What did comparison between 4 tests show? What was the goal and results?

  • Speech in noise test, method of constants
  • WIN test uses 70 of the original 200 NU6 words
  • Presented in MT babble (3 males, 3 females)
    • Level of MT babble fixed at 80 dB SPL 
    • Level of speech varied from 104 dB SPL to 80 dB SPL, in 4 dB decrements (Modified method of constants)
      • Produces SNRs of + 24 dB -> 0 dB
      • 5 words presented at each level
    • Stop presentation when listener misses all words at one SNR
      • SNR (50%) calculated as 26 – (#correct) (.8)  
        • .8 is attenuation step size (4 dB/# words/step, or 5)
        • 26 is starting SNR of 24 + correction of ½ step (or 2 dB)
    • Mean SNR for normal listeners:
      • 3.6 dB at 70 dB SPL presentation level
      • 4.8 dB at 90 dB SPL
      • 90th percentile for normal: 6 dB or less
    • Mean SNR for hearing-impaired listeners
      • 12.6 dB at 70 dB SPL presentation level
      • 12.5 dB at 90 dB SPL
    • Wilson et al. (2007): compared performance of normal hearing and hrg imp L’s on 4 speech-in-noise tests:
      • SIN, HINT, Q-SIN, WIN
      • Goal: determine which tests distinguished performance of the 2 groups best
      • Results: Q-SIN and WIN distinguished performance of the 2 groups by 8-10 dB
        • Difference in perf bet 2 grps was 4-6 dB with SIN and HINT
          • Smaller separation between groups on SIN and HINT


Describe the AzBio Sentence test. What kind of test is it? What was it developed for? How many lists and sentences per list? What are characteristics of the sentences? How are they presented? What is the task? What did Spehar and Dorman observe and find?

  • Speech test in noise, fixed signal to noise ratio
  • Developed to test listeners with CIs in noise
    • Wanted a sentence test with list equivalence in noise
    • Wanted a test that was unfamiliar to listeners; multiple lists
    • Wanted to avoid floor and ceiling effects with CI listeners  
  • Characteristics of test:
    • 15 lists of 20 sentences/ea (commercial recording)
      • Sentence length: 3-12 words, no proper nouns
      • No restrictions on complexity, vocabulary, or phonemic content
      • Sentences include contemporary adult topics and social ideas
      • 4 talkers: 2 males and 2 females
    • Presented in Q or Noise (10-talker babble)
    • Task: repeat back sentence heard; all words in sentence are scored as correct or incorrect
  • Spehar & Dorman (2004) - tested 15 listeners with CIs; mean score was 68% correct
    • Observed similar average performance for individuals across lists and minimum between-subject variability with NH and CI listeners
    • But: combined data in Q and N (SNR = +5 or + 10 dB) conditions 


Describe the validation study of AzBio by Schafer, Pogue, and Milrany. WHat did they examine? Who were the subjects? What was the level of presentation? What were the results for the NH? What were the results for the CI listeners? How did the CI and NH data compare? What was the conclusion?

  • Schafer, Pogue, & Milrany (2012)
    • Examined list equivalency of AzBio Sentence test in noise for NH and CI listeners, listening at fixed SNR
    • S’s: NH: n = 14; CI: n = 12
    • NH L’s: tested at SNR = 0 dB & -3dB; CI L’s: SNR = +10 dB
      • Signal level for both groups: 73 dB SPL; loudspeaker presentation
    • Results: NH L’s: scores ranged from 21-92% (0 dB SNR) and from  21-78% (-3 dB SNR); no ceiling or floor effects
      • Signif ME of list, SNR, and interaction; Lists 1, 6, 7, 12, & 14 - different from others
      • Re-analysis with these lists removed - remaining 10 lists were equivalent
    • CI Listener performance at + 10 dB SNR: scores ranged from 18.2 –97.8%
      • Also saw signif ME of list, but with the 5 lists removed (1, 6, 7, 12, 14)
        • no ME of list.
    • Comparing data of NH and CI L’s:
      • NH L’s (tested at 0 dB SNR) > CI L’s (at + 10 dB SNR)
      • NH L’s (tested at -3 dB SNR) = CI L’s (at + 10 dB SNR)
      • Also saw limited variation in performance with either group
    • Conclusion: 10 lists of the AzBio Sentence Test can be used as a valid, reliable, sensitive measure for clinical and research purposes.


Discuss the Perceptually Robust English Sentence Test. What was the goal of the test development? What were the stimuli? How did key words vary? Who were they recorded by? How is it administered? Is it reliable? Are ists equivalent?

  • PRESTO (Perceptually Robust English Sentence Test) (Gilbert, Tamati & Pisoni, 2013) 
    • Goal of test development: need for a test that
      • reveals individual differences among listeners
      • reflects linguistic and cognitive skills needed to understand speech
        • See how well people adapt to multiple talkers (like group conversations)
    • Stimuli: TIMIT Sentences (Texas Instruments/MIT)
      • Sentence lists (n = 19) with 18 sentences/list, with 76 keywords/list
        • Keywords varied in familiarity and word frequency, but each list was balanced for these issues
        • Sentences differ in length and syntactic structure
      • Recorded by multiple talkers, ♀s and ♂s; different dialects and ages
        • no talker was repeated in any single list; no sentence is repeated across lists
      • Administration: listener types in sentences on computer; self-paced
        • Presented at multiple SNRs in 6-talker babble (3 males, 3 females) 
        • Keywords correct are scored
      • Initial study examined test-retest reliability, which was high (r = .92)
      • Faulkner et al. (2015): different test conditions (babble, CI simulations) influence list equivalence;  ≈ 10 lists are equivalent in babble at 0 dB SNR


Describe the articulation index theory. What does the theory predict? What is it based on?

  • Articulation Index (AI) Theory
    • Fletcher & Steinberg (1947; Fletcher & Galt, 1950; ANSI, 1969; ANSI, 1997 –SII)
    • Theory predicts speech recognition performance based on:
      • Long-term spectrum of speech
      • Audibility of the speech spectrum at each of 20 bands important for speech
      • Long-term spectrum of noise
      • SNR at each of the 20 bands
      • Transfer function for a particular speech material
      • Predict score if you know patient's threshold, presentation lvel and noise level


What is the long term speech spectrum? What is the articulation index?

  • Range of speech sounds that contribute to intelligibility (30 dB)
  • If entire speech range is audible and undistorted, then speech understanding is optimal (>95% for phonemes) 
  • Ex: spectrally shaped noise = speech spectrum
  • Results in an identical change in SNR across the entire speech spectrum
  • Speech understanding ↑above 0% when SNR ↑ above –15 dB SNR (for nonsense syllables)
  • Figure shows SNRs at -10, 0, and + 10 dB SNR AI is the proportion of speech that remains visible (audible) in the presence of competing noise

A image thumb