Date post: | 21-Dec-2015 |
Category: |
Documents |
Upload: | baldric-johns |
View: | 213 times |
Download: | 0 times |
Relationship between perception of spectral ripple and speech
recognition in cochlear implant and vocoder listeners
L.M. Litvak, A.J. Spahr, A.A. Saoji, and G.Y. Fridman
L.M. Litvak, A.J. Spahr, A.A. Saoji, and G.Y. Fridman
Relationship between perception of spectral ripple
and speech recognition in cochlear implant and vocoder listeners
Frequency (Hz)
Cochlear Implant usersVariability
Why?
Can we explain this variability by testing normal listeners?Where an explanation arise?Start from the beginning – the stimulation point.
• Cochlear Implant: Auditory Nerve– Electric fields, overlap = distortion
• Normal Listener: Basilar membrane– Auditory filters, spread = decreased spectral resolution
Stimulation point
Frequency (Hz)Spe
ctra
l Lev
el
Frequency (Hz)Spe
ctra
l Lev
el
How do we change the spectral resolution in normal listeners?
Vocoder Simulations• Vocoder – electronic device that synthesizes speech• Vocoder Simulations
– Reduces spectral information to 15 channels• Mimics CI processing
– Drop-off varied 5 – 40 dB/octave• Mimics variable spectral resolution
How do we measure the changes in normal listeners?
Frequency (Hz)
Perception of Spectral RippleHow well can we represent spectral information in speech?
+ =
Frequency (Hz)
Spe
ctra
l Lev
el
Spectral modulation threshold• Spectral modulation threshold (SMT)
– measure of spectral resolution– measures the spectral ripple perception
Will the varied spectral resolution demonstrate the same variability seen in CI word recognition scores?
L.M. Litvak, A.J. Spahr, A.A. Saoji, and G.Y. Fridman
Relationship between perception of spectral ripple
and speech recognition in cochlear implant and vocoder listeners
Methods• 25 CI users, 10 normal listeners• Normal listeners
– Vocoder simulations• Speech
– Separated in 15 bands– Multiplied by noise– Change rate of drop-off of noise spectrum
» Varies spread
• Tested for recognition of vowels and consonants
• Compare word recognition scores
Frequency (Hz)
Results: Vowels
Results: Consonants
Results: Consonants
Primarily spectral cues
Primarily Temporal / Amplitude
Results
• Normal listeners– SMT increase = decrease in word recognition
scores (WRS)– Decrease in WRS similar to CI listeners with similar
SMTs
• Variability in spread (due to SMT increase) of neural activity largely accounts for variability in CI users scores.
Conclusions
• Main Finding:– Same slope between CI and normal listeners
• Spectral resolution = explanation of variability in CI users
• Subsidiary Finding:– Differences between vowels and consonants
• Temporal cues
Questions
• Additional Factors– Age of subjects
• Alternative explanations– Other cues besides temporal cues– Frequency to place alignment problem
• Central plasticity
Confusion Matrix: Vowels
Confusion Matrix: Consonants