Vowel formant discrimination in highfidelity speech by hearing-impaired listeners. Diane Kewley-Port, Chang Liu (also University at Buffalo,) T. Zachary Burkle Indiana University, SPHS Presented at the Acoustical Society of America Meeting, Austin, TX, Nov. 11, 2003. Thanks to SPL Lab members • • • • Larry Humes (Investigator) Maureen Coughlin (Audiologist, ABD) Kelley Anderson (Research Assistant) Bill Mills (Programmer) Formant Discrimination • Just noticeable difference between standard vowel and one with shifted formant. • Psychophysical procedures to determine thresholds formant frequency, DF (Hz). • For 10+ years, experiments have systematically varied conditions, phonetic context, F0, noise etc. • Purpose: Examine formant thresholds for hearing-impaired listeners (HI) in nearly natural speech, including sentences High-Fidelity Speech • To preserve naturalness, use STRAIGHT (Kawahara et al., 1999) synthesis • Stimulus Samples for word “bad” – Sentence – Word (standard vowel) – Word (10% F1 increment, NH, optimal listening Weber Fraction = 1.5%) Formant Thresholds Hi-Fi NH Delta F (Hz) 150.0 ISO Hi-Fi 125.0 Word Hi-Fi 100.0 Sent Hi-Fi 190% 160% 75.0 50.0 25.0 0.0 ih1 eh1 uh1 ae1 uh2 Vowel formant eh2 ae2 ih2 Experimental Factors for HI study • Formant Frequency: / I E Q Ã / F1 & F2 • Audibility: 70 dB SPL partial vs. 95 dB SPL fully • Linguistic Context: isolated vowels, words, sentences • Sent + ID task: Sentence discrimination only vs. Sentence discrimination + ID Hearing Impaired Listeners • 21 – 55 years old, N = 5 • Mild – moderate, high-frequency loss Procedures • • • • Day 1 Screening Days 2-4 Training Days 5-23 Testing Linguistic Context (ISO, Word, Sent) and Sent + ID blocks randomized daily • 95 vs. 70 dB SPL levels fixed each day Summary Threshold Results Factor Significant • Formant Frequency (8) Yes • Audibility (70 vs 95) No • Linguistic Context (ISO, Word, Sent) Yes • Sent + ID task No • Explain with figures Isolated Vowels Hi-fi 300.0 ISO HI 70 ISO HI 95 Delta F (Hz) 1) Formant Frequency ISO NH 70 250.0 200.0 150.0 100.0 2)Audibility 50.0 0.0 ih1 eh1 uh1 ae1 uh2 eh2 ae2 ih2 Vowel formant Sentences Hi-fi 300.0 Sent 95 250.0 Delta F (Hz) Sent 70 200.0 150.0 100.0 50.0 0.0 ih1 eh1 uh1 ae1 uh2 Vowel formant eh2 ae2 ih2 3) Linguistic Context. Thresholds different • Post-hocs, only DF word < DF Sent • Why? Linguistic Context Hi-Fi HI 300 ISO 250 Word Sent Delta F (Hz) 200 150 100 50 0 ih1 eh1 uh1 ae1 uh2 Vowel formant eh2 ae2 ih2 Reversal, DF Iso > DF Word Linguistic Context 95 dB Hi-Fi HI 300 ISO Delta F (Hz) 250 Word 200 150 100 50 0 ih1 eh1 uh1 ae1 uh2 Vowel formant eh2 ae2 ih2 Comparison HI to NH (Hi-Fi) HI vs. NH 70 dB SPL Hi-Fi 300 250 Word NH Word HI Delta F (Hz) 200 Sent NH 150 Sent HI 100 50 0 ih1 eh1 uh1 ae1 uh2 Vowel formant eh2 ae2 ih2 Thresholds Hi-Fi vs. Synthetic Speech • Richie, Kewley-Port, & Coughlin (2003) reported DF for isolated formant synthesized vowels (Syn) for HI • Liu & Kewley-Port (2003) report for NH no difference Hi-Fi and Syn for isolated vowels and words • Predict that thresholds for our Hi-Fi vowels same as Syn vowels from Richie et al. •Hi-Fi elevated by 150% Hi-Fi vs. Syn Isolated Vowels Soft 300.0 Hi-Fi NH Delta F (Hz) 250.0 Syn HI Hi-Fi HI 200.0 150.0 100.0 50.0 0.0 ih1 eh1 uh1 ae1 uh2 Vowel formant eh2 ae2 ih2 Summary • Formant discrimination by HI significantly effected by – Formant Frequency – Linguistic Context – Speech quality (Hi-Fi harder) • Surprising Hi-Fi threshold comparisons – Thresholds for softer sentences better than louder – Thresholds for words better than isolated vowels Baseline Thresholds Normal Hearing Listeners (NH) Formant Synthesized (Syn) Female Isolated (ISO) Vowels F1 & F2 Four Vowels: / I E Q Ã / Formant Thresholds Syn NH 100 90 80 Delta F (Hz) • • • • 70 60 50 40 30 20 10 0 ih1 eh1 uh1 ae1 uh2 Vowel formant ISO ae2 eh2 ih2 Linguistic Context Syn Formant Thresholds Syn NH 100 90 250% 80 Delta F (Hz) 70 170% 60 50 40 30 20 10 0 ih1 eh1 uh1 ae1 uh2 Vowel Sent formant ISO 22 C VC ae2 eh2 ih2 Added ID Task Thresholds with or without ID Task Hi-Fi HI 300 250 Threshold (Hz) 200 95 95 70 70 150 100 50 0 0 500 1000 1500 Formant fre que ncy (Hz) 2000 2500 dB, dB, dB, dB, Sen. sen+ID sen sen+ID Audibility versus Pathology • Vowels fully audible 70 dB NH, 95 dB HI DF2 elevated by 200 % Fully Audible HI vs. NH Word Hi-Fi 300 250 NH 70 Delta F (Hz) 200 HI 95 150 100 50 0 ih1 eh1 uh1 ae1 uh2 Vowel formant eh2 ae2 ih2 Liste n e r V ariab ility at 70 d B fo r W o rd 400.0 300.0 D elta F (H z) B AC BC C 200.0 C LM D LN JR M 100.0 0.0 ih1 eh1 uh1 ae1 uh2 V ow elFrom ant eh2 ae2 ih2