If It's Good Enough for You, It's Good Enough for Me: Transferability of Audio Sufficiencies across Models

David A. Kelly; Hana Chockler

cs.SDApr 3, 2026

If It's Good Enough for You, It's Good Enough for Me: Transferability of Audio Sufficiencies across Models

Authors: David A. Kelly, Hana Chockler

Organizations: King’s College London UK

Abstract

In order to gain fresh insights about the information processing characteristics of different audio classification models, we propose transferability analysis. Given a minimal, sufficient signal for a classification on a model $f$ , transferability analysis asks whether other models accept this minimal signal as having the same classification as it did on $f$ . We define what it means for a sufficient signal to be transferable and perform a large study over $3$ different classification tasks: music genre, emotion recognition and deepfake detection. We find that transferability rates vary depending on the task, with sufficient signals for music genre being transferable $\approx26\%$ of the time. The other tasks reveal much higher variance in transferability and reveal that some models, in particular on deepfake detection, have different transferability behavior. We call these models `flat-earther' models. We investigate deepfake audio in more depth, and show that transferability analysis also allows to us to discover information theoretic differences between the models which are not captured by the more familiar metrics of accuracy and precision.

Figures & tables

Figure 1. The partitioning of ‘disgust’ (a) into sufficient signal (b), sufficient and necessary (complete) signal (c) and

1

-complete signal (d). A

1

-complete signal is sufficient and necessary for ‘disgust’ and also has the same confidence as the original signal (a). We investigate how transferable these signals and classifications are across different models.

Panel of Figure 2 — Figure 2. Graphical representation of an audio depth- $2$ causal model with $n$ frequencies as input, and classification as outcome. The internal node $f_{i}$ introduces the network $f$ as a black-box.

			Wav2Vec2		Hubert	DistHubt.	Whisp.
Model	Size (M)	Acc.	VC 1	VC 2	VC 3	VC 4	VC 5	Avg.
VC 1	$86.6$	$0.89$	–	0.123	0.278	0.289	0.212	0.23
VC 2	$23.4$	$0.87$	0.051	–	0.077	0.101	0.789	0.25
VC 3	$8.3$	$0.91$	0.17	0.073	–	0.184	0.189	0.15
VC 4	$23.4$	$0.9$	0.175	0.123	0.192	–	0.159	0.16
VC 5	$23.4$	$0.82$	0.288	0.016	0.062	0.077	–	0.11

Table 1.

\alpha

-transferability of voice emotion sufficient (top) and complete (bottom) signals across models. Size is in millions of parameters. We have also indicated which type of backbone of each model.

			DistHubt			Whisp.	Trans.
Model	Size (M)	Acc.	MC 1	MC 2	MC 3	MC 4	MC 5	Avg.
MC 1	$23.7$	$0.82$	–	0.171	0.408	0.248	0.24	0.27
MC 2	$94.6$	$0.86$	0.327	–	0.238	0.248	0.265	0.27
MC 3	$23.5$	$0.89$	0.47	0.218	–	0.253	0.19	0.28
MC 4	$8.31$	$0.91$	0.315	0.211	0.351	–	0.305	0.29
MC 5	$86.6$	$0.89$	0.146	0.21	0.174	0.261	–	0.2

Table 2.

\alpha

-transferability of music genre sufficient (top) and complete (bottom) signals across models; “Size” is in millions of parameters. In general, complete signals are more easily transferable across models.

				Wav2Vec2
Dataset	Model	Size (M)	Acc.	SP 1	SP 2	SP 3	Avg.
	SP 1	$300$	$0.93$	–	0.859	0.782	0.83
	SP 2	$94.6$	$0.99$	0.192	–	0.882	0.54
ASV	SP 3	$94.6$	$0.99$	0.354	0.254	–	0.3
Spoof2019	SP 1	$300$	$0.93$	–	0.764	0.664	0.71
	SP 2	$94.6$	$0.99$	0.856	–	0.766	0.81

Table 3.

\alpha

-transferability of spoof sufficient (top) and complete (bottom) signals across models on ASVspoof2019 and ITW. “Size” is in millions of parameters. Highlighted are average transferability rates which are unusual.

	ASVspoof		ITW
	Class		Class
Model	real	fake	real	fake
SP 1	0.996	0.47	1.0	0.64
SP 2	0.6	0.97	0.57	0.95
SP 3	0.47	0.99	0.52	1.0

Table 4. Degree of compositionality for real and fake classifications across 3 models and ASVspoof2019 (left) and ITW (right).

Figure 3. Power spectral density of real and fake data on ASVSpoof2019 (top

2

rows) and ITW (bottom two rows). The different between SP 1 and the other models in clear.

Figure 4. Spectral entropy across real and fake sufficiencies on ASVSpoof2019 (top) and ITW (bottom) and

3

models.

Model	Sufficient		Complete		Inverse STOI
	real	fake	real	fake	real	fake
SP 1	$0.99$	$0.99$	$0.99$	$0.98$	$0.14$	$0.15$
SP 2	$0.99$	$0.98$	$0.99$	$0.98$	$0.05$	$0.06$
SP 3	$0.99$	$0.98$	$0.99$	$0.98$	$0.02$	$0.01$
SP 1	$0.99$	$0.99$	$0.99$	$0.98$	$0.58$	$0.63$
SP 2	$0.99$	$0.99$	$0.99$	$0.98$	$0.03$	$0.07$

Table 5. Levenshtein ratio of sufficient and complete signals using an ASR model as oracle. While all models show virtually identical fidelity on sufficient and complete signals, there is considerable difference on the intelligibility of the inverse signal: the signal remaining after the complete signal has been removed. SP 1 is again an outlier, which a great deal of intelligible signal remaining, unlike the other models.

Audio Classification

Audio classification assigns recordings to categories such as speech, music, animal calls, acoustic scenes, and health-related events.

See all 147 papers

How Class Ontology and Data Scale Affect Audio Transfer Learning
Mar 26, 2026Manuel Milling, Andreas Triantafyllopoulos, Alexander Gebhard +2Transfer Learning Audio Classification
CHI – Chair of Health Informatics, Technical University of Munich, Munich, Germany · MCML – Munich Center for Machine Learning, Munich, Germany · MDSI – Munich Data Science Institute, Munich, Germany +1
The Perceived Fragility of Explanations in Audio Models: Manipulation of Attribution with Unchanged Predictions
Jun 12, 2026Piotr Kitłowski, Dominik Wiącek, Mateusz ModrzejewskiAudio Deepfake Detection Adversarial Attacks
Faculty of Electronics and Information Technology, Warsaw University of Technology, Warsaw, Poland.
Unsupervised Domain Adaptation for Audio Deepfake Detection with Modular Statistical Transformations
Mar 9, 2026Urawee Thani, Gagandeep Singh, Priyanka SinghAudio Deepfake Detection Audio Deepfake Detection Robustness

Mar 26, 2026J/K move · Enter open · S save

How Class Ontology and Data Scale Affect Audio Transfer Learning

Manuel Milling, Andreas Triantafyllopoulos, Alexander Gebhard, Simon Rampp+1

CHI – Chair of Health Informatics, Technical University of Munich, Munich, Germany · MCML – Munich Center for Machine Learning, Munich, Germany · MDSI – Munich Data Science Institute, Munich, Germany+1

Transfer Learning Audio Classification

Open paper AskPDF arXiv

Transfer learning is a crucial concept within deep learning that allows artificial neural networks to benefit from a large pre-training data basis when confronted with a task of limited data. Despite its ubiquitous use and clear benefits, there are still many open questions regarding the inner workings of transfer learning and, in particular, regarding the understanding of when and how well it works. To that extent, we perform a rigorous study focusing on audio-to-audio transfer learning, in which we pre-train various model states on (ontology-based) subsets of AudioSet and fine-tune them on three computer audition tasks, namely acoustic scene recognition, bird activity recognition, and speech command recognition. We report that increasing the number of samples and classes in the pre-training data both have a positive impact on transfer learning. This is, however, generally surpassed by similarity between pre-training and the downstream task, which can lead the model to learn comparable features.

Abstract

Audio Classification

Explore similar work

How Class Ontology and Data Scale Affect Audio Transfer Learning

The Perceived Fragility of Explanations in Audio Models: Manipulation of Attribution with Unchanged Predictions

Unsupervised Domain Adaptation for Audio Deepfake Detection with Modular Statistical Transformations