1 / 9

distance, m (log scale)

-25 o. 0 o. +25 o. left. left. left. right. right. right. room 1l08. room 155. room 157. 12. 12. 12. 12. 12. 12. 12. 12. 12. 12. 12. C 50 , dB. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 2.5. 2.5. 2.5. 2.5. 2.5. 2.5. 10. 10. 10. 10. 10. 10. corridor. L-shaped.

sadie
Download Presentation

distance, m (log scale)

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. -25o 0o +25o left left left right right right room 1l08 room 155 room 157 12 12 12 12 12 12 12 12 12 12 12 C50, dB 0 0 0 0 0 0 0 0 0 0 0 2.5 2.5 2.5 2.5 2.5 2.5 10 10 10 10 10 10 corridor L-shaped distance, m (log scale)

  2. Some further data on ‘wide-band scn’ contexts scn is speech → noise operation is performed on the wide-band signal (unlike the vocoder manipulations) reverse polarity of a randomly-selected half of the samples, then apply a speech-shaping filter ‘noise before’ conditions: room impulse response (BRIR) applied after the scn operation in this condition of Watkins, 2005, expt. 4: no effect of test-word’s reverb.

  3. This experiment similar except: ‘to click on ‘ part of the context stays as speech in ‘noise before’ conditions monaural (left ear) stressed (slow) test-words and context speech looks at the effect of a context-length (685 ms) gap preceding the test-word 4 observations per stimulus from each of 6 listeners L-shaped room, as before

  4. no gap gap • compensation effect with speech • gap reduces the effect, but it’s not entirely eliminated • effect is marginal or absent in noise (before) conditions • no influence from ‘to click on’ 10 speech 5 0 category boundary, step test dist.=10. m test dist.=.32 m noise (before) 10 5 0 context’s distance, m .32 .32 10. 10.

  5. Temporal envelopes of context and test-word in an auditory (gammatone) filter, fc= 4.2 kHz • as reverb increases, mean increases • but this effect is if anything more apparent with the noise context 10. m 0.32 m speech amplitude re. max 1 1 noise (before) 0 0 1 0 0 1 time, s

  6. Is temporal structure of reverb pattern important? forwards reversed • 8-band vocoder, one channel: • if scn comes after the BRIR operation • temporal structure is scrambled • e.g. as measured by autocorrelation: • consider a single reflection • scn flattens its combfilter spectrum speech → → scn → → gammatonegammatone BRIR

  7. unprocessed speech 4-,8-band mismatched 4-,8-band matched mismatched, both groups group 1 context test word matched, group 2 category boundary, step matched, group 1 test dist.=10. m test dist.=.32 m 10 10 group 2 category boundaries higher than in mismatched 5 5 could well be an inverse filter effect if even bands have more info about the [t] … 0 0 .32 .32 .32 10. 10. 10. context’s distance, m

  8. Inverse-filter effects with wideband (smother) temporal envelope? forwards reversed scn → → scn → → gammatonegammatone • replace speech input with wideband scn • 8-band vocoder, one channel: BRIR

  9. Band importance experiments using compensation effect, rather effects of reverb on the test-word.

More Related