Julius Richter

Downstream-Task-Aware Unified Source Separation

Task-aware unified source separation (TUSS) enables a single model to handle diverse separation tasks by conditioning on input prompts. However, conventional TUSS does not account for downstream task requirements, such as whether the enhanced speech will be used for human listening or automatic speech recognition (ASR)...

Dual-Geometry Manifolds for Few-shot RIR Prediction

Few-Shot Room Impulse Response Interpolation in Latent Domains

Transcript-Supervised Post-Training of Generative Speech Enhancement on Real Recordings via Reinforce Adjoint Matching

We adapt Reinforce Adjoint Matching (RAM), a reward-based post-training method, to generative speech enhancement (SE). Starting from a pretrained SE model, RAM tilts the model's conditional distribution toward outputs with higher reward. During training, the current model generates enhanced speech on-policy, evaluates...

Anomalous Sound Detection Meets Noise-Aware Self-Supervised Learning

Noise-aware self-supervised learning (NA-SSL) models for noise-aware anomalous sound detection (NA-ASD) are introduced and the effectiveness of the proposed approach is demonstrated across three base SSL models (BEATs, EAT, and Dasheng), both with and without discriminative fine-tuning.

Technical Report for MERL's Real-TSE Challenge Submission

MERL's submission to the Real-TSE Challenge achieved first place in the second track, demonstrating the critical importance of high-quality data preparation and observing that DNSMOS and speaker similarity are susceptible to over-optimization.