Setup

• Open-fitting hearables in both ears of a GRAS 45BB-12 KEMAR head-and-torso simulator
• Four outer mics (two entrance, two concha; #1–#4), one inner error mic at the right eardrum (#5), plus the right-ear outer receiver as secondary source
• Desired speech from 0° (VCTK “p361_005”)
• Two babble-noise sources at 45° and 255° (NOISEX-92)
• All signals: 5 s long, 16 kHz sampling
• Inner error mic. SNR set to –5 dB
• Secondary path estimate: measured impulse response from outer receiver → inner error mic.
• Convolution of anechoic impulse responses (from Denk et al.) with source signals

Fig. 2: (a) Illustration of the open-fitting hearable. (b) Acoustic scenario with one desired speech source at 0^o and two babble noise sources at 45^o and 255^o.

Results

Fig. 3: Noise reduction, intelligibility-weighted spectral distortion, intelligibility-weighted SNR improvement, narrowband PESQ improvement, and ESTOI improvement for conventional ANC (left column, µ = 0), the hard-constrained SSANC (right column, µ → ∞), and the proposed soft-constrained SSANC for different values of the trade-off parameter µ.

Audio samples