Few-Shot Calibration for Sim-to-Real Single-Channel Speaker Distance Estimation

2026-09-24

AI Summary

This study addresses the poor sim-to-real transferability of simulation-trained distance estimation models and the scarcity of annotated real-world data. To overcome these challenges, we propose a post-processing calibration mapping framework based on ranking consistency. By revealing that distance errors are fundamentally constrained by ordinal ranking rather than absolute accuracy, we design a linear correlation-based checkpoint selection criterion and a variant without hard-decision shrinkage. This approach corrects scale and bias using only minimal samples without requiring model retraining. Our method significantly improves the accuracy of single-channel distance estimation in real-world scenarios and validates the effectiveness of selecting synthetic-domain pretrained models via linear correlation, thereby establishing a new paradigm for cross-domain few-shot calibration.

Abstract

Speaker distance estimators are trained almost exclusively on simulated room acoustics, because real recordings annotated with the true talker-to-microphone distance are scarce. We show that models trained this way transfer poorly. On three real corpora we evaluate, simply predicting the average distance of the corpus is more accurate than any learned model. Then, we ask how few labelled real utterances are needed to make a frozen, synthetic-trained estimator useful, and study post-hoc calibration maps that rescale its output without gradients or retraining. An analysis of the achievable error shows that what the calibration is not limited by the absolute accuracy of the estimator, but how well it orders utterances by distance, since a constant bias or a wrong output scale is removed exactly by the calibration itself. Balancing this against the cost of estimating each coefficient from few samples yields a criterion that accounts for which map wins on which corpus and at which annotation budget, together with a shrinkage variant that requires no hard decision. Our findings suggest selecting synthetic checkpoints by linear correlation with true distances rather than by absolute error. Code, datasets, and analysis are available at https://github.com/michaelneri/audio-distance-estimation.

Problem

• speaker distance estimation
• sim-to-real transfer
• few-shot calibration
• room acoustics

Innovation

• Few-shot calibration
• Sim-to-real transfer
• Speaker distance estimation
• Post-hoc calibration maps
• Shrinkage