# Tiny is not small enough: High-quality, low-resource facial animation models through hybrid knowledge distillation

arXiv: https://arxiv.org/abs/2507.18352 · HTML: https://mlsanity.com/paper/2507.18352/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Placed in Speech and audio distillation, one of 12 directions in Knowledge distillation.
- Builds on 25 of its 59 references that are in the map.
- First posted on arXiv 24 Jul 2025.

## Placed in

- Knowledge distillation / [Speech and audio distillation](https://mlsanity.com/distill/speech-and-audio-distillation/)

## Builds on (most cited first, 25 of 25)

- [Adam: A Method for Stochastic Optimization](https://mlsanity.com/paper/1412.6980/) (2014)
- [PyTorch: An Imperative Style, High-Performance Deep Learning Library](https://mlsanity.com/paper/1912.01703/) (2019)
- [Distilling the Knowledge in a Neural Network](https://mlsanity.com/paper/1503.02531/) (2015)
- [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://mlsanity.com/paper/2006.11477/) (2020)
- [HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units](https://mlsanity.com/paper/2106.07447/) (2021)
- [FitNets: Hints for Thin Deep Nets](https://mlsanity.com/paper/1412.6550/) (2014)
- [Knowledge Distillation: A Survey](https://mlsanity.com/paper/2006.05525/) (2020)
- [Paying More Attention to Attention: Improving the Performance of Convolutional Neural Networks via Attention Transfer](https://mlsanity.com/paper/1612.03928/) (2016)
- [Capture, Learning, and Synthesis of 3D Speaking Styles](https://mlsanity.com/paper/1905.03079/) (2019)
- [Learning Deep Representations with Probabilistic Knowledge Transfer](https://mlsanity.com/paper/1803.10837/) (2018)
- [CodeTalker: Speech-Driven 3D Facial Animation with Discrete Motion Prior](https://mlsanity.com/paper/2301.02379/) (2023)
- [Emotional Speech-Driven Animation with Content-Emotion Disentanglement](https://mlsanity.com/paper/2306.08990/) (2023)
- [Distil-Whisper: Robust Knowledge Distillation via Large-Scale Pseudo Labelling](https://mlsanity.com/paper/2311.00430/) (2023)
- [DiffPoseTalk: Speech-Driven Stylistic 3D Facial Animation and Head Pose Generation via Diffusion Models](https://mlsanity.com/paper/2310.00434/) (2023)
- [FaceDiffuser: Speech-Driven 3D Facial Animation Synthesis Using Diffusion](https://mlsanity.com/paper/2309.11306/) (2023)
- [Imitator: Personalized Speech-driven 3D Facial Animation](https://mlsanity.com/paper/2301.00023/) (2022)
- [Heterogeneous Knowledge Distillation using Information Flow Modeling](https://mlsanity.com/paper/2005.00727/) (2020)
- [Efficient Region-Aware Neural Radiance Fields for High-Fidelity Talking Portrait Synthesis](https://mlsanity.com/paper/2307.09323/) (2023)
- [FaceTalk: Audio-Driven Motion Diffusion for Neural Parametric Head Models](https://mlsanity.com/paper/2312.08459/) (2023)
- [Audio- and Gaze-driven Facial Animation of Codec Avatars](https://mlsanity.com/paper/2008.05023/) (2020)
- [Media2Face: Co-speech Facial Animation Generation With Multi-Modality Guidance](https://mlsanity.com/paper/2401.15687/) (2024)
- [FaceXHuBERT: Text-less Speech-driven E(X)pressive 3D Facial Animation Synthesis Using Self-Supervised Speech Representation Learning](https://mlsanity.com/paper/2303.05416/) (2023)
- [LaughTalk: Expressive 3D Talking Head Generation with Laughter](https://mlsanity.com/paper/2311.00994/) (2023)
- [Online Multi-Granularity Distillation for GAN Compression](https://mlsanity.com/paper/2108.06908/) (2021)
- [Rig Inversion by Training a Differentiable Rig Function](https://mlsanity.com/paper/2301.09567/) (2023)

## Built on by (most cited first, 2 of 2)

- [UMo: Unified Sparse Motion Modeling for Real-Time Co-Speech Avatars](https://mlsanity.com/paper/2605.14731/) (2026)
- [TokTalk: Expressive Real-time Facial Animation from Audio-LLM Tokens](https://mlsanity.com/paper/2605.31294/) (2026)

## Abstract (arXiv)

The training of high-quality, robust machine learning models for speech-driven 3D facial animation requires a large, diverse dataset of high-quality audio-animation pairs. To overcome the lack of such a dataset, recent work has introduced large pre-trained speech encoders that are robust to variations in the input audio and, therefore, enable the facial animation model to generalize across speakers, audio quality, and languages. However, the resulting facial animation models are prohibitively large and lend themselves only to offline inference on a dedicated machine. In this work, we explore on-device, real-time facial animation models in the context of game development. We overcome the lack of large datasets by using hybrid knowledge distillation with pseudo-labeling. Given a large audio dataset, we employ a high-performing teacher model to train very small student models. In contrast to the pre-trained speech encoders, our student models only consist of convolutional and fully-connected layers, removing the need for attention context or recurrent updates. In our experiments, we demonstrate that we can reduce the memory footprint to up to 3.4 MB and required future audio context to up to 81 ms while maintaining high-quality animations. This paves the way for on-device inference, an important step towards realistic, model-driven digital characters.