---
title: Bone-Conduction Diffusion (BCDM-DC-L)
url: https://www.emergentmind.com/topics/bone-conduction-diffusion-bcdm-dc-l
type: topic
---

# Bone-Conduction Diffusion (BCDM-DC-L)

The Bone-conduction Diffusion model, large variant (BCDM-DC-L), is a multimodal speech enhancement architecture that integrates air- and bone-conducted microphone inputs via a conditional diffusion generative framework. Designed to address severe performance degradation in single-channel speech enhancement systems under extreme noise, BCDM-DC-L explicitly leverages the complementary, noise-robust characteristics of bone-conducted speech to guide the denoising process in the spectral domain through a mathematically grounded stochastic differential equation (SDE) mechanism [2601.12354].

## 1. Mathematical Foundation

BCDM-DC-L models the clean speech short-time Fourier transform (STFT) slice $x_0 \in \mathbb R^D$, the noisy air-conducted STFT $y$, and the bone-conducted STFT $y_c$. The generative process is defined as a continuous-time SDE—specifically, a drift-diffusion process:

\[
dx_t = f(x_t, y)\,dt + g(t)\,dW_t, \quad t \in [0, T]
\tag{1}
\]

where the drift term
\[
f(x_t, y) = \gamma (y - x_t), \quad \gamma > 0
\tag{2}
\]
pulls the process toward the noisy air-conducted mixture, and the diffusion scheduling is 

\[
g(t) = \sigma_{\min} \left(\frac{\sigma_{\max}}{\sigma_{\min}}\right)^{t} \sqrt{2\ln\left(\frac{\sigma_{\max}}{\sigma_{\min}}\right)}
\tag{3}
\]
with $\sigma_{\min} = 0.05$, $\sigma_{\max} = 0.5$. $W_t$ is standard Wiener noise.

The corresponding reverse-time SDE is
\[
dx_t = \big[ -f(x_t, y) + g(t)^2 \nabla_{x_t} \log p_t(x_t \mid y) \big] dt + g(t) d\overline{W}_t
\tag{4}
\]
where the "score" $\nabla_{x_t} \log p_t(x_t \mid y)$ is approximated by a neural network $s_\theta(x_t, y, y_c, t) \approx \nabla_{x_t} \log p_t(x_t \mid y, y_c)$. Discretization over $T$ steps yields the forward transition:

\[
q(x_i \mid x_{i-1}, y) = \mathcal N (x_i;\, x_{i-1} + \gamma (y - x_{i-1})\Delta t,\, g_i^2 I)
\tag{5}
\]
and the reverse (denoising) process is modeled as
\[
p_\theta(x_{i-1} \mid x_i, y, y_c) = \mathcal N (x_{i-1};\, \mu_\theta(x_i, y, y_c, i),\, \sigma_i^2 I)
\tag{6}
\]
where $\mu_\theta$ is constructed via an Euler–Maruyama step utilizing the learned score $s_\theta$.

Training minimizes the (conditional) Fisher divergence between true and model score:

\[
\mathcal{L}(\theta) = \mathbb{E}_{t \sim \mathcal{U}[0,1], x_0, y, y_c, x_t \sim q} \left\|\, s_\theta(x_t, y, y_c, t) - \nabla_{x_t}\log q(x_t\mid x_0, y)\, \right\|_2^2
\tag{7}
\]

In practice, the equivalent "noise-prediction" parameterization is adopted: 
\[
\mathcal{L}(\theta) = \mathbb{E}\left[\|\varepsilon - \varepsilon_\theta(x_t, y, y_c, t)\|^2\right]
\tag{8}
\]
No additional variational bounds are used.

## 2. Model Architecture

The BCDM-DC-L architecture is based on a multi-resolution NCSN++ U-Net, comprising symmetric encoder-decoder paths. The "large" (L) variant adopts channel widths $[128, 256, 512, 1024]$ and ResNet blocks of depth 2, with approximately 67.4 million parameters.

**Bone-conduction conditioning (DC)** is realized via a separate BigGAN-style ResNet downsampling encoder for $y_c$ featuring depth 2 and matching channel sizes. Decoder-stage integration concatenates: (i) main decoder skip-connection features, (ii) the previous decoder output, and (iii) the bone-conduction encoder's features; a $1 \times 1$ convolution reduces channel dimensionality post-concatenation.

#### Data flow outline:

- $x \leftarrow$ noisy mixture spectrogram
- $x_t \leftarrow$ forward SDE state sampled from $x$
- For $i = T \downarrow 1$:
  - $\text{score} \leftarrow \text{ScoreNet}(x_t, y=x, y_c, i)$
  - $x_{t-1} \leftarrow x_t + [-\gamma(y-x_t) + g_i^2 \cdot \text{score}]\Delta t + g_i \zeta,\; \zeta\sim \mathcal{N}(0,I)$
  - $x_t \leftarrow x_{t-1}$
- Output $x_0$ (complex STFT of enhanced speech)

## 3. Training Methodology

The model is trained with the loss
\[
\mathcal L(\theta) = \mathbb{E}_{t,x_0,y,y_c,\varepsilon}\left[\|\varepsilon - \varepsilon_\theta(x_t,y,y_c,t)\|^2_2\right]
\tag{9}
\]
where $x_t = \mu_t(x_0, y) + \sigma_t \varepsilon$ with $(\mu_t, \sigma_t)$ provided directly from SDE analysis.

Optimization employs Adam (learning rate $10^{-4}$; $\beta=(0.9,0.999)$), batch size 8, with exponential moving average parameter tracking (decay 0.999) and $T=60$ diffusion steps per sample.

## 4. Dataset and Preprocessing

Data is sourced from ABCS: 42 hours, 100 speakers, with aligned air- and bone-conducted speech. Noise is synthetically added to air signals using CHiME3, training SNR sampled uniformly in $[-5,\,20]$ dB and testing on SNRs $-10,\, -5,\, 0,\, 5,\, 15$ dB. STFT parameters are: window size 510, hop 128, yielding 256 frequency bins and 256 time frames post-processing, resulting in $256 \times 256$ complex-valued inputs. Each STFT channel is per-utterance zero-mean, unit-variance normalized.

## 5. Inference and Sampling Procedure

At inference, a Predictor–Corrector sampler is used, cycling through:

- Predictor: Euler–Maruyama step using $s_\theta$.
- Corrector: Langevin step; $x \leftarrow x + \alpha \nabla_x \log p_\theta(x \mid y, y_c) + \sqrt{2\alpha} \xi,\; \xi \sim \mathcal{N}(0,I)$.

Typically, $N=60$ steps are used, but step counts $10$–$60$ are supported. Step size $\alpha$ is adapted to maintain sampler SNR. The model achieves real-time feasibility (e.g., 256×256 input, $N=60$ in 150 ms with a single GPU); real-time/causal deployment is projected possible with 8–16 steps ($\sim$30 ms/frame). The small model variant achieves competitive results (PESQ) with as few as 20 steps, corresponding to roughly 40 score evaluations.

## 6. Empirical Performance and Model Analysis

BCDM-DC-L demonstrates consistent superiority over prior single- and multimodal baselines (BiNet, DCCRN) in multiple metrics:

| SNR (dB) | POLQA | PESQ | ESTOI |
|----------|-------|------|-------|
| –10      | 2.44  | 2.02 | 0.76  |
| +15      | 3.70  | 3.25 | 0.94  |

Relative improvements over the strongest baseline are $+0.20$ to $+0.40$ in POLQA, $+0.2$ to $+0.4$ in PESQ, and $+0.05$ to $+0.10$ in ESTOI across all SNRs evaluated.

Ablation studies contrast "DC" (bone-conduction conditioning) versus "IC" (independent conditioning): DC (67.4 M parameters) outperforms IC (65.6 M) by $0.05$ to $0.10$ PESQ but exhibits slower performance scaling as the number of reverse steps increases. Smaller BCDM-DC-S variants surpass DCCRN at $N\approx10$ steps and BiNet at $N \approx 20$.

## 7. Significance and Context

BCDM-DC-L demonstrates the effectiveness of integrating bone-conducted signals through conditional diffusion modeling, leveraging the intrinsic noise-resilience of bone transmission and the generative flexibility of score-based SDEs. These results substantiate the potential of multimodal SDE-guided diffusion models to robustly enhance speech in adverse acoustic regimes, setting a new reference for quantitative performance in the multimodal speech enhancement domain [2601.12354]. The explicit SDE-based fusion of modalities, and efficient inference via predictor–corrector samplers, positions this approach for further adoption both in batch and real-time applications where robust speech quality under noise is requisite.

Source: https://www.emergentmind.com/topics/bone-conduction-diffusion-bcdm-dc-l