Effect of Carrier Bandwidth on Understanding Mandarin Sentences in Simulated Electric-Acoustic Hearing

Feng Wang, Jing Chen, Fei Chen2021 · DOI

TL;DR

This paper investigates how the bandwidth of electrical stimulation, specifically using either noise or pure-tone carriers, impacts speech understanding in simulated electric-acoustic hearing (EAS). The findings show that with limited electrical channels, EAS significantly boosts intelligibility, and surprisingly, a broader noise carrier can outperform a narrow pure-tone carrier for certain configurations.

What Problem Does This Paper Address?

For individuals with high-frequency hearing loss, cochlear implants (CIs) offer a way to restore hearing by electrically stimulating the auditory nerve. However, many of these individuals retain some low-frequency hearing. Combined electric-acoustic stimulation (EAS) aims to leverage this residual low-frequency hearing by providing it acoustically, while the CI handles the high frequencies electrically. This approach has shown a "combined-stimulation advantage," improving speech perception compared to electric-only stimulation.

Despite the promise of EAS, research has often explored scenarios with a relatively large number of electrical channels. This paper addresses a critical gap: understanding the benefits and nuances of EAS when only a limited number of electrical channels (specifically, one or two) are used. Furthermore, it delves into the effect of the carrier signal's bandwidth in the electrical stimulation component, comparing noise-based carriers (which can simulate broader excitation) with pure-tone carriers (which can simulate more focused excitation). This is important because the choice of carrier and its bandwidth can influence how speech information is encoded and perceived by the auditory system.

Methods

The researchers employed a vocoder simulation approach, presenting synthesized speech stimuli to normal-hearing listeners. This method allows for controlled manipulation of acoustic and electric processing without the variability inherent in testing actual cochlear implant users.

The core of their methodology involved three main processing conditions:

  • Acoustic-only (A-only): Speech was low-pass filtered to simulate the residual low-frequency hearing.
  • Electric-only (E-only): Speech was processed through a vocoder to simulate CI stimulation. This involved band-pass filtering the speech into multiple channels, extracting the temporal envelope of each channel, and then using this envelope to modulate either a white noise carrier (Noise-Vocoder, NV) or a pure-tone carrier (Tone-Vocoder, TV). The number of electrical channels (N) was varied (N=1 or N=2).
  • Combined Electric-Acoustic Stimulation (EAS): This condition merged the A-only low-pass filtered speech with the E-only vocoded speech.

A key aspect of their investigation was the manipulation of carrier bandwidth. The NV condition, using white noise, inherently represents a broader carrier bandwidth. The TV condition, using pure tones, was intended to simulate a more focused excitation, especially when combined with specific filtering. The paper also varied the low-pass cutoff frequency for the acoustic portion (250 Hz and 500 Hz).

Datasets

The speech material used for the experiment was extracted from the Mandarin Hearing in Noise Test (MHINT) database. This database consists of sentences designed for speech recognition testing. The corpus includes 24 lists, each containing 10 sentences with 10 keywords per sentence. The sentences were spoken by a male native Mandarin Chinese speaker.

Key Insights and Ideas

  • Limited Electrical Channels Still Yield Significant EAS Benefit: Even with just one or two electrical channels, combining them with low-frequency acoustic information dramatically improves speech intelligibility compared to electric-only stimulation. This highlights the power of EAS even in resource-limited CI scenarios.
  • Carrier Bandwidth Matters, Especially with Few Channels: The type of carrier signal used in the electrical component has a notable impact. While pure-tone carriers (TV) are often thought to provide more focused information, the study found that with a single electrical channel (N=1), the broader white noise carrier (NV) actually led to better intelligibility.
  • Interaction Between Acoustic and Electric Components: The study suggests that the presence of low-frequency acoustic information can influence the perceived advantage of different carrier types in the electrical domain. The typical advantage of TV over NV in electric-only simulations was diminished or even reversed in some EAS conditions.
  • Focus vs. Spread in Electrical Stimulation: The paper implicitly explores the trade-off between focused electrical stimulation (simulated by TV) and broader excitation (simulated by NV). The results suggest that for very limited electrical channels, broader excitation might be more beneficial, potentially by providing more robust envelope information or better spectral coverage.

Key Results

  • Acoustic-only stimulation with low-pass filtering (250 Hz or 500 Hz cutoff) resulted in low intelligibility scores.
  • Combined EAS significantly improved intelligibility scores to over 86.0% when compared to acoustic-only stimulation.
  • Electric-only stimulation: Tone-vocoded (TV) stimuli generally yielded higher intelligibility scores than noise-vocoded (NV) stimuli when N=1 or N=2.
  • EAS with N=1 electrical channel:
    • A significant effect of LP cutoff frequency and carrier signal type was observed.
  • NV stimuli were significantly more intelligible than TV stimuli at both 250 Hz and 500 Hz LP cutoff frequencies. This is a key finding, as it contrasts with typical electric-only results.
  • EAS with N=2 electrical channels:
    • A significant effect of LP cutoff frequency was observed, but the effect of carrier signal type was not significant.
  • There were no significant performance differences between paired TV and NV speech. This suggests that with more electrical channels, the advantage of the noise carrier diminishes.
  • In both N=1 and N=2 EAS conditions, the combined stimulation advantage was observed compared to electric-only stimulation.

Limitations and Future Work

  • Simulation vs. Real-World: The study uses normal-hearing listeners and vocoder simulations. While this controls for many variables, it doesn't fully replicate the experience of individuals with cochlear implants, whose auditory systems may process information differently.
  • Limited Electrical Channels: The focus on N=1 and N=2 electrical channels is a strength for exploring specific scenarios but might not generalize to CI users with more channels.
  • Carrier Bandwidth Definition: While NV and TV carriers are used to represent different bandwidths, the exact spectral characteristics and their interaction with the auditory system could be further explored.

The authors suggest that future work could investigate the impact of carrier bandwidth in EAS with a larger number of electrical channels and explore how different carrier types interact with other CI processing parameters. Further research into the perceptual mechanisms underlying the observed trade-offs between carrier type and channel number in EAS would also be valuable.

Summary updated

BibTeX
@misc{wang2021effect,
  author = {Feng Wang and Jing Chen and Fei Chen},
  title = {{Effect of Carrier Bandwidth on Understanding Mandarin Sentences in Simulated Electric-Acoustic Hearing}},
  year = {2021},
  doi = {10.21437/interspeech.2021-24},
}

Loading related content...

Figure