# Breaking the Quality--Intelligibility Trade-off in Streaming Target Speaker Extraction via Deep-Feature-Anchored Preference Optimization

arXiv: https://arxiv.org/abs/2607.10191 · HTML: https://mlsanity.com/paper/2607.10191/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 12 of its 16 references that are in the map.
- First posted on arXiv 11 Jul 2026.

## Builds on (most cited first, 12 of 12)

- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Robust Speech Recognition via Large-Scale Weak Supervision](https://mlsanity.com/paper/2212.04356/) (2022)
- [WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing](https://mlsanity.com/paper/2110.13900/) (2021)
- [DNSMOS: A Non-Intrusive Perceptual Objective Speech Quality metric to evaluate Noise Suppressors](https://mlsanity.com/paper/2010.15258/) (2020)
- [LibriMix: An Open-Source Dataset for Generalizable Speech Separation](https://mlsanity.com/paper/2005.11262/) (2020)
- [Wespeaker: A Research and Production oriented Speaker Embedding Learning Toolkit](https://mlsanity.com/paper/2210.17016/) (2022)
- [FunCodec: A Fundamental, Reproducible and Integrable Open-source Toolkit for Neural Speech Codec](https://mlsanity.com/paper/2309.07405/) (2023)
- [SpEx+: A Complete Time Domain Speaker Extraction Network](https://mlsanity.com/paper/2005.04686/) (2020)
- [TSELM: Target Speaker Extraction using Discrete Tokens and Language Models](https://mlsanity.com/paper/2409.07841/) (2024)
- [WeSep: A Scalable and Flexible Toolkit Towards Generalizable Target Speaker Extraction](https://mlsanity.com/paper/2409.15799/) (2024)
- [LauraTSE: Target Speaker Extraction using Auto-Regressive Decoder-Only Language Models](https://mlsanity.com/paper/2504.07402/) (2025)
- [StarTSE: Towards Streaming Target Speaker Extraction via Chunk-wise Interleaved Splicing of Autoregressive Language Model](https://mlsanity.com/paper/2604.19635/) (2026)

## Abstract (arXiv)

Generative streaming models for Target Speaker Extraction (TSE) commonly exhibit a quality--intelligibility trade-off, wherein naive optimization for perceptual audio quality tends to degrade speech intelligibility, and conversely. We reveal that this trade-off arises not from the constraints of streaming architectures, but from an inappropriate choice of optimization anchor. Directly optimizing against audio quality metrics induces catastrophic reward hacking, where content critical to pronunciation and intelligibility is systematically erased to maximize a proxy score. To break this bottleneck, we propose two complementary improvements: an enlarged Conformer convolution kernel for richer local spectro-temporal modeling, and WavLM-anchored Direct Preference Optimization (DPO) fine-tuning strategy. DPO preference pairs are ranked by WavLM cosine similarity, a deep acoustic feature encoding both phonetic structure and speaker identity, providing an optimization anchor that resists hacking. Under a 560 ms streaming chunk size, the proposed method achieves a 10.9% relative intelligibility improvement (word error rate: 0.138 to 0.123), with marginal simultaneous gains in audio quality and speaker similarity.