# Mixed-Precision Information Bottlenecks for On-Device Trait-State Disentanglement in Bipolar Agitation Detection

arXiv: https://arxiv.org/abs/2605.03039 · HTML: https://mlsanity.com/paper/2605.03039/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 21 of its 42 references that are in the map.
- First posted on arXiv 4 May 2026.

## Builds on (most cited first, 21 of 21)

- [Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning](https://mlsanity.com/paper/1506.02142/) (2015)
- [Domain-Adversarial Training of Neural Networks](https://mlsanity.com/paper/1505.07818/) (2015)
- [Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation](https://mlsanity.com/paper/1308.3432/) (2013)
- [Searching for MobileNetV3](https://mlsanity.com/paper/1905.02244/) (2019)
- [WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing](https://mlsanity.com/paper/2110.13900/) (2021)
- [ECAPA-TDNN: Emphasized Channel Attention, Propagation and Aggregation in TDNN Based Speaker Verification](https://mlsanity.com/paper/2005.07143/) (2020)
- [CMSIS-NN: Efficient Neural Network Kernels for Arm Cortex-M CPUs](https://mlsanity.com/paper/1801.06601/) (2018)
- [Unsupervised Speech Decomposition via Triple Information Bottleneck](https://mlsanity.com/paper/2004.11284/) (2020)
- [VQMIVC: Vector Quantization and Mutual Information-Based Unsupervised Speech Representation Disentanglement for One-shot Voice Conversion](https://mlsanity.com/paper/2106.10132/) (2021)
- [Cross-speaker emotion disentangling and transfer for end-to-end speech synthesis](https://mlsanity.com/paper/2109.06733/) (2021)
- [TinySpeech: Attention Condensers for Deep Speech Recognition Neural Networks on Edge Devices](https://mlsanity.com/paper/2008.04245/) (2020)
- [Effective and Efficient Mixed Precision Quantization of Speech Foundation Models](https://mlsanity.com/paper/2501.03643/) (2025)
- [DiEmo-TTS: Disentangled Emotion Representations via Self-Supervised Distillation for Cross-Speaker Emotion Transfer in Text-to-Speech](https://mlsanity.com/paper/2505.19687/) (2025)
- [Edge-ASR: Towards Low-Bit Quantization of Automatic Speech Recognition Models](https://mlsanity.com/paper/2507.07877/) (2025)
- [ZSVC: Zero-shot Style Voice Conversion with Disentangled Latent Diffusion Models and Adversarial Training](https://mlsanity.com/paper/2501.04416/) (2025)
- [Towards One-bit ASR: Extremely Low-bit Conformer Quantization Using Co-training and Stochastic Precision](https://mlsanity.com/paper/2505.21245/) (2025)
- [Speaker and Style Disentanglement of Speech Based on Contrastive Predictive Coding Supported Factorized Variational Autoencoder](https://mlsanity.com/paper/2409.03520/) (2024)
- [Towards Better Disentanglement in Non-Autoregressive Zero-Shot Expressive Voice Conversion](https://mlsanity.com/paper/2506.04013/) (2025)
- [Breaking Resource Barriers in Speech Emotion Recognition via Data Distillation](https://mlsanity.com/paper/2406.15119/) (2024)
- [Provable Speech Attributes Conversion via Latent Independence](https://mlsanity.com/paper/2510.05191/) (2025)
- [Learnable Pulse Accumulation for On-Device Speech Recognition: How Much Attention Do You Need?](https://mlsanity.com/paper/2603.16922/) (2026)

## Abstract (arXiv)

Continuous monitoring of bipolar disorder agitation via voice biomarkers requires disentangling stable speaker traits from volatile affective states on resource-constrained edge devices. We introduce MP-IB, the first framework to treat mixed-precision quantization as an information bottleneck for clinical trait-state separation. The core insight is that numerical precision itself controls capacity: an FP16 trait head (1,024 bits) encodes speaker identity, while an INT4 state head (128 bits) captures agitation, yielding 8x information asymmetry without adversarial training. We augment this with Dynamic Precision Scheduling and Multi-Scale Temporal Fusion. On Bridge2AI-Voice (N=833, 4 sessions/participant, strict speaker-independent CV), MP-IB achieves rho = 0.117 (95% CI: [0.089, 0.145], p=0.003 vs. chance), outperforming 94M-parameter WavLM-Adapter with in-domain SSL continuation (rho = -0.042), beta VAE disentanglement (rho = 0.089), and hand-crafted prosody (rho = 0.031) by 2.8--15.9 points absolute. Zero-shot transfer to CREMA-D achieves AUC=0.817. Identity leakage is suppressed to near-random (EER=0.42, MIA-AUC=0.52). End-to-end latency is 23.4 ms with a 617 KB footprint, enabling real-time monitoring on sub 20 dollar devices.