The effect of dichotic processing on the perception of binaural cues

Akiko Amano-Kusumoto, Justin M. Aronoff, Motokuni Itoh, Sigfrid D. SoliInterspeech 2012

TL;DR

This paper investigates how dichotic processing, which splits audio frequency bands between ears, affects speech perception and sound localization in normal-hearing individuals. They found that while dichotic processing can degrade binaural cues crucial for spatial hearing, a "partial dichotic" approach—combining diotic and dichotic filtering—can mitigate these negative effects, offering a promising avenue for hearing aid technology.

What Problem Does This Paper Address?

Hearing-impaired individuals often struggle with understanding speech in noisy environments due to reduced spectral resolution. Dichotic processing algorithms aim to improve this by presenting different frequency bands to each ear, potentially reducing "upward spread of masking." However, a significant challenge is that this splitting of frequency information can disrupt binaural cues (interaural time and level differences) that are vital for spatial hearing, including sound localization and the ability to separate speech from background noise. This paper seeks to quantify this disruption and explore whether a modified dichotic approach can offer the benefits of improved spectral resolution without sacrificing essential binaural cue information.

Methods

The researchers employed an eight-channel filter bank to process auditory stimuli. They compared three filter configurations:

  • Diotic: All frequency bands are presented to both ears. This serves as a baseline.
  • Dichotic: Odd-numbered frequency bands are sent to one ear, and even-numbered bands to the other. This is the core dichotic processing being investigated.
  • Partial Dichotic: The lowest and highest two frequency bands are presented diotically, while the middle four bands are presented dichotically. This hybrid approach aims to preserve binaural cues in the frequency ranges where they are most critical (low frequencies for ITDs, high frequencies for ILDs) while still benefiting from dichotic processing in the mid-range.

Two experiments were conducted with normal-hearing subjects:

  1. Hearing in Noise Test (HINT): This measured speech reception thresholds (SRTs) in various noise conditions (co-located, spatially separated).
  2. Source Azimuth Localization Test (SALT): This assessed the accuracy of sound source localization.

Head-related transfer functions (HRTFs) were used to simulate realistic sound field testing using headphones. Statistical analyses, including percentile-t bootstrap repeated measures ANOVAs with trimmed means, were used to account for potential outliers and non-normality.

Datasets

Not discussed in the paper. The paper describes the generation of stimuli using HINT sentences and speech-shaped noise, but does not refer to pre-existing datasets for training or evaluation.

Key Insights and Ideas

  • Dichotic processing is detrimental to binaural cues when spatial separation is present: When speech and noise were spatially separated, dichotic processing significantly worsened both speech intelligibility and sound localization accuracy compared to diotic processing. This confirms the hypothesis that splitting frequency bands disrupts the binaural cues necessary for spatial hearing.
  • Partial dichotic processing offers a compromise: The "partial dichotic" configuration, which kept the lowest and highest frequency bands diotic, significantly reduced the negative impact on binaural cues. It performed comparably to diotic processing in terms of localization and speech intelligibility in noise, suggesting it can preserve crucial spatial information.
  • Spatial separation is key: When speech and noise were co-located (no spatial separation), dichotic processing did not significantly affect performance. This implies that normal-hearing listeners can integrate speech information effectively even when it's divided across ears, as long as there's no competing sound source to localize or separate spatially.
  • Trade-off between spectral resolution and binaural cues: The study highlights a fundamental trade-off. While dichotic processing might offer benefits in spectral resolution (reducing masking), it comes at the cost of binaural cue disruption. The partial dichotic approach attempts to balance these, but the authors suggest further investigation is needed to understand how this balance affects hearing-impaired listeners.

Key Results

  • HINT (Speech in Noise):
    • With spatially separated noise (NR/NL), dichotic processing led to significantly worse SRTs (higher dB needed for intelligibility) compared to diotic processing.
  • Partial dichotic processing did not significantly differ from diotic processing in SRTs, indicating it preserved intelligibility.
  • When noise was co-located (NF), there was no significant effect of filter configuration on SRTs.
  • SALT (Sound Localization):
    • Dichotic processing resulted in significantly larger RMS errors (worse localization accuracy) compared to diotic processing.
  • Partial dichotic processing did not significantly differ from diotic processing in localization accuracy.
  • The "diotic benefit" (improvement from diotic over dichotic) was significantly larger for the dichotic configuration than the partial dichotic one, reinforcing that dichotic processing was more detrimental.

Limitations and Future Work

  • Normal-Hearing Subjects: The study was conducted on normal-hearing individuals. The authors suggest that the observed trade-offs and benefits might differ for hearing-impaired listeners, particularly those with reduced spectral resolution.
  • Trade-off for Hearing-Impaired Listeners: The paper notes that while partial dichotic processing preserves binaural cues, it might also reduce the potential "dichotic benefit" that hearing-impaired listeners could gain from improved spectral resolution. Future work should investigate this trade-off in hearing-impaired populations.
  • Number of Bands: The study used an eight-channel filter bank. Further research could explore the impact of different numbers of channels or filter designs on the observed effects.

Summary updated

BibTeX
@inproceedings{amanokusumoto2012effect,
  author = {Akiko Amano-Kusumoto and Justin M. Aronoff and Motokuni Itoh and Sigfrid D. Soli},
  title = {{The effect of dichotic processing on the perception of binaural cues}},
  booktitle = {Interspeech},
  year = {2012},
}
Figure