Differentiable Modal Synthesis for Physical Modeling of Planar String Sound and Motion Simulation

While significant advancements have been made in music generation and differentiable sound synthesis within machine learning and computer audition, the simulation of instrument vibration guided by physical laws has been underexplored. To address this gap, we introduce a novel model for simulating the spatio-temporal motion of nonlinear strings, integrating modal synthesis and spectral modeling within a neural network framework. Our model leverages physical properties and fundamental frequencies as inputs, outputting string states across time and space that solve the partial differential equation characterizing the nonlinear string. Empirical evaluations demonstrate that the proposed architecture achieves superior accuracy in string motion simulation compared to existing baseline architectures. The code and demo are available online.

Paper

References (66)

Scroll for more · 38 remaining

Similar papers

Peer review

Reviewer Hh9k6/10 · confidence 2/52024-07-09

Summary

The paper presents a differentiable model that can synthesize musical string sound and simulate motion based on physical properties. The method uses finite-difference time-domain (FDTD) solver to obtain numerical solutions and take them as ground truth. Then a differentiable pipeline with neural network components is used to map physics properties to output waveform. Experimental results show that it achieves better performance than Modal synthesis and DDSPish baselines.

Strengths

- To my best knowledge, it is the first differentiable method that can generate the musical string sound and motion from physics properties. - The proposed method is more efficient and accurate than baselines.

Weaknesses

- The proposed method relies FDTD solutions as ground truth and I am wondering what will be the gap between the simulated solutions versus real-world audio data. - Following the previous question, there might not be too many real-world data available for learning and usually material properties are unavailable. Is it possible to infer these properties through the differentiable pipeline? - It needs ablation studies for using two MLPs for modulation layers, as well as a MLP for mode estimator. - There are references typos in L135 and L229.

Questions

See questions in weakness

Rating

6

Confidence

2

Soundness

2

Presentation

2

Contribution

2

Limitations

Authors mentioned limitations in the main text.

Reviewer ygxN5/10 · confidence 3/52024-07-13

Summary

This paper proposes Differentiable Modal Synthesis for Physical Modeling (DMSP), which is a neural-model-based method to predict the vibration of nonlinear strings. DMSP takes the physical parameters for the differentiable equation as the input, predicts the mode frequencies and AM/FM effects, and finally outputs the waveform of vibration.

Strengths

S1: This paper is among the first attempts to study neural audio synthesis using physical properties, which makes a meaningful contribution in this sub-area. S2: Section 2 provides some background which is helpful for understanding the context of the proposed method.

Weaknesses

W1: Some method details are not introduced with sufficient clarity. In the loss section, the detailed mathematical forms for pitch loss and mode frequency loss are not explained. Is the pitch loss L2 loss, L1 loss, or in other forms? What type of regularization is used for the mode frequency loss? Also, since there are typically many mode frequencies for a linear/nonlinear system, it is unclear how many mode frequencies the mode estimator predicts. Is it a fixed number or a variable number? If it is a variable number, how did the neural model output a variable number of mode frequencies? W2: As the authors have also discussed, the mode predictor turns out to be not functioning appropriately based on the results in Table 2 (the big gap between DMSP and DMSP-N). Since the mode predictor is a major component of the proposed system, the design of the proposed pipeline is not well-justified. W3: It is generally unclear what advantages the proposed physics-driven sound synthesis paradigm has over neural-based generative models, such as AudioLM or diffusion models. The latter has been shown to synthesize high-quality sounds, music, and speech. Does the proposed paradigm have higher quality, lower computational overhead, greater generalizability, or better in other aspects than the latter paradigm? W4: (Minor) There are many empty references, e.g., line 135, line 229, and wrong references, e.g. Table 1 is line 243 which should probably be Table 2. Also, in line 243, ‘the ablation of the mode information is studied..’ Shouldn’t this table be the main result? Calling it an ablation result rather than the main result may cause some confusion, especially when Table 3 shows another ablation study result. W5: (Minor) The problem statement in 3.1 is a bit confusing to me. For example, it is unclear to me why the initial condition $u_0$, is an element of $\mathcal{U}$. $u_0$ is a function on $\Omega \times \{0\}$, whereas $\mathcal{U}$ is a set of $(x, t)$ pairs. Also, in line 138, it seems that the mapping $\mathcal{S}$ is not only a function of $\mathcal{P}$, but also a function of initial conditions. Why is the latter omitted?

Questions

Please see ‘weaknesses’.

Rating

5

Confidence

3

Soundness

3

Presentation

1

Contribution

3

Limitations

Limitations are minimally discussed in Section 5. It would be nice to extend the discussion to include the failure of the mode predictor, the computational complexity of DMSP-N, etc.

Reviewer gawP7/10 · confidence 2/52024-07-13

Summary

A computational framework that can approximate the motion of nonlinear strings is proposed. The implementation is differentiable, so one can train then neural nets in the framework as usual.

Strengths

- The paper is written nicely such that I could follow the basic discussions even if I am not familiar with the topics of speech and audio. It seems to include all the information required in this kind of paper, such as technical background, model description, loss description, and experiments. - The empirical performance of the method is clearly superior to the reasonable baselines.

Weaknesses

I have only minor comments as follows. Basically no need to respond to them. - In Table 1, the computational complexity part does not consider the computation needed for learning unknown parameters. The same kind of complexity analysis may be difficult for that part, but at least, the fact that such a training procedure happens for some of the methods should be noted somewhere close. - Lines 137 and 138: there seems to be confusion about the definition of the initial condition and/or the operator $\mathcal{S}$. First, $u_0 \in \mathcal{U}$ sounds strange if $\mathcal{U}$ is the space of functions $\Omega \times [0,\infty) \to \mathbb{R}$ because $u_0$ does not take the time $\in [0,\infty)$ as the argument. Second, the domain of $\mathcal{S}$ is the product of $\mathcal{P}$ and the space of initial conditions, isn't it? - The role of the noise decoder is unclear. Is it important in the given experiments where the data are purely from the simulator?

Questions

I don't have major questions.

Rating

7

Confidence

2

Soundness

3

Presentation

4

Contribution

3

Limitations

Limitations are simply stated in the conclusion section, which sound reasonable.

Area Chair P6or2024-08-12

Dear Reviewer ygxN, As the discussion between the reviewers and authors is coming to an end, could you please respond to the authors to confirm whether your concerns have been addressed? Thanks! AC

Area Chair P6or2024-08-12

Dear Reviewer Hh9k, As the discussion between the reviewers and authors is coming to an end, could you please respond to the authors to confirm whether your concerns have been addressed? Thanks! AC

Reviewer Hh9k2024-08-12

Thanks for the response. I appreciate the clarification and additional ablation studies. My concerns are mostly addressed and I am happy to raise my score.

Reviewer ygxN2024-08-12

Thanks for the response. Most of my concerns are addressed and I am happy to raise the score.

Reviewer gawP2024-08-12

Thanks for the authors' rebuttal, it further clarified the discussion. I would maintain my originally positive score.

Program Chairsdecision2024-09-25

Decision

Accept (poster)

© 2026 NYSGPT2525 LLC