The effect of spectral smearing on the identification of pureF₀intonation contours in vocoder simulations of cochlear implants

Abstract

Objectives: Performance of cochlear implant (CI) users on linguistic intonation recognition is poorer than that of normal-hearing listeners, due to the limited spectral detail provided by the implant. A higher spectral resolution is provided by narrow rather than by broad filter slopes. The corresponding effect of the filter slope on the identification of linguistic intonation conveyed by pitch movements alone was tested using vocoder simulations. Methods: Re-synthesized intonation variants of naturally produced phrases were processed by a 15-channel noise vocoder using a narrow (40 dB/octave) and a broad (20 dB/octave) filter slope. There were three different intonation patterns (rise/fall/rise-fall), differentiated purely by pitch and each associated to a different meaning. In both slope conditions as well as a condition with unprocessed stimuli, 24 normally hearing Dutch adults listened to a phrase, indicating which of two meanings was associated to it (i.e. a counterbalanced selection of two of the three contours). Results: As expected, performance for the unprocessed stimuli was better than for the vocoded stimuli. No overall difference between the filter conditions was found. Discussion and conclusions: These results are taken to indicate that neither the narrow (20 dB/octave) nor the shallow (40 dB/octave) slope provide enough spectral detail to identify pure F 0 intonation contours. For users of a certain class of CIs, results could imply that their intonation perception would not benefit from steeper slopes. For them, perception of pitch movements in language requires more extreme filter slopes, more electrodes, and/or additional (phonetic/contextual) cues.

The perception of emotion and focus prosody with varying acoustic cues in cochlear implant simulations with varying filter slopes

“…The literature reviewed above suggests that, similar to segmental perception, prosodic pitch (i.e., intonation) perception benefits from better frequency selectivity in the form of steeper filter slopes. However, whereas for segmental identification scores reached asymptote at 40 dB/octave (Litvak et al, 2007), performance for intonation perception was still at chance for 40 dB/octave (van de Velde et al, 2015), despite using the same number of channels (though some other vocoding parameters differed between the studies). Given the results of those studies, we hypothesize that, given comparable tasks, intonation perception requires greater channel independence, perhaps as realized by means of electrode configuration or steeper filter slopes, than segmental perception, because intonation perception relies more heavily on spectral versus temporal information relative to segmental perception.…”

“…Slopes with a shallow roll-off overlap each other more than those with a steep roll-off, resulting in more spectral smearing. Moreover, even with steep analysis filters, spectral smearing is also induced by overlapping neuron areas stimulated by adjacent electrodes (Tang et al, 2011), a factor represented by means of the synthesis filter in vocoder simulations. Using vocoder simulations of CIs, this study aims to find the theoretically optimal filter slope for the perception of a specific aspect of speech in which pitch plays a central role (i.e., prosody).…”

Recognition and cortical haemodynamics of vocal emotions-an fNIRS perspective

“…Just as accuracy scores vary between listeners, the ability to make use of cues conveyed by intact acoustic features to compensate for uninformative F0 cues may differ between listeners (e.g., listeners presented full-spectrum speech or CI-simulations and CI listeners presented full-spectrum speech recognise vocal emotions more accurately when overall intensity cues are intact, i.e., not normalised. Gilbers et al (2015) suggest that informative overall intensity cues do not improve recognition of vocal emotions perceived through a CI or make use of speech rate cues to recognise vocal emotions, the majority of studies suggest listeners cannot make use of speech-rate cues to identify emotions conveyed in speech with uninformative F0 cues (Gilbers et al, 2015;Luo, 2016;Van de Velde, 2017). Combined, these studies highlight the lack of consensus regarding listeners' 66 abilities to make use of intensity or speech-rate cues in the presence of uninformative F0 cues.…”

“…Evidence exists that increased reliance on intensity cues may support recognition of emotions in the absence of F0 cues to emotional prosody (Chapter 3;Luo et al, 2007), although intensity cues fe, 2017). Other studies suggest that listeners cannot use intensity cues to compensate for uninformative F0 cues when identifying vocal emotions (Gilbers et al, 2015;Luo, to make use of speech-rate cues to recognise emotions in speech in which F0 cues are uninformative, while other studies indicate that whether or not listeners increase their reliance on speech rate when parsing vocal emotions, they are unable to make use of speech-rate cues to recognise emotions in speech with uninformative F0 cues (Chapter 3; Gilbers et al, 2015;Luo, 2016;Van de Velde, 2017). Further, speech rate, of itself, 2017).…”