mlsanityData to 7 Oct 2026

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

arXiv 2605.03039 ↗First posted 4 May 2026cs.LG cs.AI cs.HCcited by 0

Not on the map. Why?

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.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 21 of its 42 references that are in the map. First posted on arXiv 4 May 2026.

Builds on

21 of 42 references.

References · 21

  1. 2015Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learningcited by 4,416
  2. 2015Domain-Adversarial Training of Neural Networkscited by 3,023
  3. 2013Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computationcited by 2,031
  4. 2019Searching for MobileNetV3cited by 1,984
  5. 2021WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processingcited by 1,609
  6. 2020ECAPA-TDNN: Emphasized Channel Attention, Propagation and Aggregation in TDNN Based Speaker Verificationcited by 606
  7. 2018CMSIS-NN: Efficient Neural Network Kernels for Arm Cortex-M CPUscited by 116
  8. 2020Unsupervised Speech Decomposition via Triple Information Bottleneckcited by 87
  9. 2021VQMIVC: Vector Quantization and Mutual Information-Based Unsupervised Speech Representation Disentanglement for One-shot Voice Conversioncited by 63
  10. 2021Cross-speaker emotion disentangling and transfer for end-to-end speech synthesiscited by 21
  11. 2020TinySpeech: Attention Condensers for Deep Speech Recognition Neural Networks on Edge Devicescited by 16
  12. 2025Effective and Efficient Mixed Precision Quantization of Speech Foundation Modelscited by 5
9 more
  1. 2025DiEmo-TTS: Disentangled Emotion Representations via Self-Supervised Distillation for Cross-Speaker Emotion Transfer in Text-to-Speechcited by 5
  2. 2025Edge-ASR: Towards Low-Bit Quantization of Automatic Speech Recognition Modelscited by 5
  3. 2025ZSVC: Zero-shot Style Voice Conversion with Disentangled Latent Diffusion Models and Adversarial Trainingcited by 3
  4. 2025Towards One-bit ASR: Extremely Low-bit Conformer Quantization Using Co-training and Stochastic Precisioncited by 3
  5. 2024Speaker and Style Disentanglement of Speech Based on Contrastive Predictive Coding Supported Factorized Variational Autoencodercited by 2
  6. 2025Towards Better Disentanglement in Non-Autoregressive Zero-Shot Expressive Voice Conversioncited by 2
  7. 2024Breaking Resource Barriers in Speech Emotion Recognition via Data Distillationcited by 1
  8. 2025Provable Speech Attributes Conversion via Latent Independencecited by 1
  9. 2026Learnable Pulse Accumulation for On-Device Speech Recognition: How Much Attention Do You Need?cited by 1

Built on by

Not cited yet.