---
title: 'DISPatch: Selective Patches for Speech Enhancement'
url: https://www.emergentmind.com/topics/distilling-selective-patches-dispatch
type: topic
---

# DISPatch: Selective Patches for Speech Enhancement

Searching arXiv for the cited papers to ground the article in current records.
Distilling Selective Patches (DISPatch) is a knowledge distillation framework for speech enhancement that applies the distillation loss only to spectrogram patches where the teacher outperforms the student, as determined by a Knowledge Gap Score. In contrast to conventional response-based KD, which enforces full-spectrogram imitation, DISPatch targets regions with the largest potential for student improvement and suppresses distillation in regions where the student already performs well or where the teacher is unreliable. The framework is defined at the patch level in the time–frequency domain and is extended by Multi-Scale Selective Patches (MSSP), which uses different patch sizes for low- and high-frequency bands to account for spectral heterogeneity [2509.15922].

## 1. Motivation and problem setting

Conventional response-based KD for speech enhancement applies a distillation loss over the entire time–frequency output of the teacher. The reported motivation for DISPatch is that such full-spectrogram imitation suffers from two problems: the student may already match the teacher, or even the clean target, in many regions, and the teacher itself is not perfect and contains low-quality regions. Under those conditions, indiscriminate KD can either yield little benefit or propagate teacher errors [2509.15922].

The framework is formulated on complex spectrograms. A clean target $\mathbf{S}$, a student estimate $\hat{\mathbf{S}}^S$, and a teacher estimate $\hat{\mathbf{S}}^T$ are split into $P$ non-overlapping patches of size $2C\times N\times 1$ along the frequency axis:
$$
\{\mathbf{s}_p,\hat{\mathbf{s}}^S_p,\hat{\mathbf{s}}^T_p\}_{p=0}^{P-1}.
$$
Patch selection is therefore frequency-structured rather than image-like: the unit of distillation is a spectrogram subregion, not a full utterance or a single frame.

A common misconception is to treat DISPatch as a mechanism for selecting the teacher’s most confident patches. The formal criterion is different. Selection is based on the *difference* between student and teacher patch-wise errors relative to the clean target, so a patch is emphasized only when the teacher is demonstrably better than the student on that region [2509.15922].

## 2. Knowledge Gap Score

DISPatch defines patch-wise student and teacher errors using a flexible error metric $\varepsilon(\mathbf{s}_p,\hat{\mathbf{s}}_p)$, with examples including $\ell_1$, $\ell_2$ magnitude distance, or the DFKD objective. For each patch $p$,
$$
E^S_p \;=\;\varepsilon\bigl(\mathbf{s}_p,\hat{\mathbf{s}}^S_p\bigr),
\qquad
E^T_p \;=\;\varepsilon\bigl(\mathbf{s}_p,\hat{\mathbf{s}}^T_p\bigr).
$$
The Knowledge Gap Score (KGS) is then
$$
KGS_p \;=\; E^S_p \;-\; E^T_p.
$$

A larger $KGS_p$ indicates that the student’s error greatly exceeds the teacher’s, so distilling that patch should yield maximal improvement. The score is therefore a patch-wise estimate of actionable teacher superiority rather than a generic saliency or confidence measure [2509.15922].

This definition has two immediate consequences. First, patches where the student has already converged to low error are naturally deprioritized. Second, teacher regions with high error are also suppressed, because they reduce or reverse the gap. This suggests that DISPatch functions as a selective transfer operator that couples KD strength to local teacher–student discrepancy, rather than to teacher output magnitude alone.

## 3. Selective-patch distillation objective

After computing $KGS_p$ over all patches, DISPatch ranks patches by their scores and retains only the top $k\%$. With the indicator
$$
I_{k\%}(KGS_p)\;=\;
\begin{cases}
1 & \text{if }KGS_p\text{ is among the top }k\%,\\
0 & \text{otherwise,}
\end{cases}
$$
the distillation loss becomes
$$
\mathcal{L}_{\mathrm{DISPatch}}
\;=\;
\frac{1}{(k\%\times P)}
\sum_{p=0}^{P-1}
I_{k\%}(KGS_p)\;\mathcal{L}_{\mathrm{KD}}\bigl(\hat{\mathbf{s}}^S_p,\hat{\mathbf{s}}^T_p\bigr).
$$

In practice, the authors set $k=80$, so 80% of the patches with highest $KGS_p$ contribute to $\mathcal{L}_{\mathrm{DISPatch}}$ [2509.15922]. The selection ratio is therefore not extremely sparse; it removes the lowest-value fifth of the spectrogram patches while keeping the majority of informative regions. This distinguishes the method from top-$k$ hard-mining schemes that retain only a very small fraction of patches.

DISPatch is agnostic to the underlying KD objective $\mathcal{L}_{\mathrm{KD}}$. The overall student training loss is
$$
\mathcal{L}_{\mathrm{total}}
\;=\;
\alpha\,\mathcal{L}_{\mathrm{SE}}
\;+\;
(1-\alpha)\,\mathcal{L}_{\mathrm{DISPatch}},
$$
where $\mathcal{L}_{\mathrm{SE}}$ is the standard speech enhancement loss and $\alpha\in[0,1]$ balances the two terms. The reported setting fixes $\alpha=0.5$ throughout [2509.15922].

Operationally, one training iteration proceeds by computing teacher and student outputs on the noisy input, evaluating $\mathcal{L}_{\mathrm{SE}}$, splitting teacher output, student output, and clean target into patches, computing $E^S_p$, $E^T_p$, and $KGS_p$, selecting the top $k\%$ patches, forming $\mathcal{L}_{\mathrm{DISPatch}}$, and then backpropagating $\mathcal{L}_{\mathrm{total}}$ to update only the student. The method is thus a selective masking layer over an otherwise conventional KD pipeline.

## 4. Multi-Scale Selective Patches (MSSP)

MSSP extends DISPatch by recognizing that speech spectrograms exhibit different structures at low and high frequencies. The stated motivation is that vowel formants are fine-grained and require small patches, whereas fricatives and noise bursts are broader and require larger patches. MSSP therefore uses two patch sizes,
$$
N_{\mathrm{low}} \quad\text{for the low-frequency band}, \qquad
N_{\mathrm{high}} \quad\text{for the high-frequency band}.
$$

The framework dynamically partitions each frame into low- and high-frequency bands using the DFKD frequency adapter. The paper’s best setting for ConvTasNet and DCCRN-CL is
$$
N_{\mathrm{low}}=10,\qquad N_{\mathrm{high}}=40,
$$
so that low-frequency patches are finer than high-frequency patches [2509.15922].

Patch selection and distillation then proceed exactly as in DISPatch, but over a heterogeneous set of patch sizes. When MSSP is integrated into a frequency-dependent KD method such as DFKD, the uniform patching step is replaced by separate low/high patching, and the per-band loss formulations of DFKD are retained: phase-only loss for the low band, and amplitude-plus-phase loss for the high band [2509.15922].

The conceptual significance of MSSP is that patch granularity becomes frequency-dependent rather than globally fixed. This suggests a domain-specific adaptation of selective KD: the patching geometry is aligned with the nonuniform structure of speech spectra instead of being imposed uniformly across all bands.

## 5. Experimental configuration and reported results

The reported experiments use two datasets: DNS2020, with 500+ h speech and 100+ h noise for training and the DNS2020 challenge set for testing, and VoiceBank+DEMAND with the standard 16 kHz test set. Two model families are evaluated: ConvTasNet, with a teacher of 9.71 M parameters and 8.05 G MAC and a student of 1.75 M parameters and 0.57 G MAC, and DCCRN-CL, with a teacher of 9.22 M and 15.33 G and a student of 2.24 M and 3.84 G. Training uses 100 epochs, Adam with learning rate $1\mathrm{e}{-3}$ halved on plateau, batch size 16, gradient clip 5, STFT with Hann 32 ms and hop 8 ms, and patch selection ratio $k=80\%$ [2509.15922].

The evaluation metrics are Wideband PESQ (WB-PESQ), Narrowband PESQ (NB-PESQ), and STOI (%). For brevity, the table below reproduces the ConvTasNet student results explicitly listed for DNS2020-test and VoiceBank+DEMAND.

| Method | DNS2020-test | VoiceBank+DEMAND |
|---|---|---|
| Scratch | 2.563 / 3.122 / 96.05 | 2.585 / 3.424 / 93.44 |
| L1-KD | 2.584 / 3.135 / 96.11 | 2.584 / 3.429 / 93.57 |
| L1 + DISPatch | 2.602 / 3.165 / 96.16 | 2.614 / 3.450 / 93.79 |
| L2-KD | 2.563 / 3.127 / 95.98 | 2.556 / 3.400 / 93.37 |
| L2 + DISPatch | 2.602 / 3.157 / 96.10 | 2.603 / 3.431 / 93.39 |
| DFKD | 2.623 / 3.170 / 96.08 | 2.583 / 3.458 / 93.70 |
| DFKD + DISPatch | 2.650 / 3.208 / 96.12 | 2.589 / 3.490 / 93.81 |
| DFKD + MSSP(10/40) + DISPatch | 2.677 / 3.226 / 96.17 | 2.618 / 3.505 / 94.02 |

The reported interpretation is consistent across methods. Adding DISPatch to L1 or L2 gives consistent WB-PESQ, NB-PESQ, and STOI gains over the corresponding KD methods without DISPatch. Even DFKD benefits from DISPatch, with a reported $+0.027$ WB-PESQ gain. The full combination DFKD+MSSP(10/40)+DISPatch yields the best overall ConvTasNet results on both DNS2020 and VoiceBank+DEMAND. The paper also states that analogous gains appear for DCCRN-CL [2509.15922].

## 6. Position within the broader patch-selection literature

The acronym “DISPatch” is not unique to speech enhancement. In dataset distillation, "Foreground-Aware Dataset Distillation via Dynamic Patch Selection" defines a method that routes each image through a cropping path or a resize path according to foreground occupancy derived from Grounded SAM2, using a class-specific threshold $\tau_c=\mathrm{Quantile}(\{\alpha_i\}_{i\in\mathcal{C}_c},q)$, typically with $q=30\%$; it reports improvements over RDED, MTT, IDM, DATM, D$^4$M, G-VBSM, and SRe$^2$L across CIFAR-10, CIFAR-100, ImageNette, and ImageWoof [2601.02727].

A second image-domain method, "Efficient Dataset Distillation via Diffusion-Driven Patch Selection for Improved Generalization," also uses the name DISPatch and scores patches by the difference between conditional and unconditional diffusion-model noise-prediction losses,
$$
R(p\mid c)=\mathbb{E}_{t,\epsilon}\bigl[\ell_c(z,\epsilon,t)-\ell_{\varnothing}(z,\epsilon,t)\bigr],
$$
then enforces diversity through intra-class clustering on DIFT features; it reports, for example, about $59.4\%$ on ImageNet-1K at IPC$=50$ with a ResNet-18 teacher, compared with about $56.6\%$ for Minimax and $56.5\%$ for RDED [2412.09959].

Related patch distillation ideas also appear outside dataset distillation. In computational pathology, AFD-MIL filters diagnostically informative patches before fusion, uses an attention mechanism to retain the top-$k$ patches by attention score, and optimizes the process with a global loss; the reported results are 91.47% ACC and 94.29% AUC on Camelyon16, and 93.33% ACC and 98.17% AUC on TCGA-NSCLC [2407.19821].

These parallel usages indicate that “selective patches” has become a recurring design motif rather than a single modality-specific technique. This suggests a broader methodological pattern: patch-level selectivity is being used to suppress redundant or harmful supervision while preserving the subregions most relevant to the downstream objective. In speech enhancement, DISPatch instantiates that pattern through teacher–student error gaps in the time–frequency domain; in image distillation and MIL, analogous roles are played by foreground occupancy, diffusion-based class relevance, or attention-derived diagnostic value.

Source: https://www.emergentmind.com/topics/distilling-selective-patches-dispatch