Pseudo2Real: Task Arithmetic for Pseudo-Label Correction in Automatic Speech Recognition
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866918453941633024 |
|---|---|
| author | Lin, Yi-Cheng Liang, Yu-Hsuan Li Su, Hsuan Lin, Tzu-Quan Chen, Shang-Tse Chen, Yun-Nung Lee, Hung-yi |
| author_facet | Lin, Yi-Cheng Liang, Yu-Hsuan Li Su, Hsuan Lin, Tzu-Quan Chen, Shang-Tse Chen, Yun-Nung Lee, Hung-yi |
| contents | Robust ASR under domain shift is crucial because real-world systems encounter unseen accents and domains with limited labeled data. Although pseudo-labeling offers a practical workaround, it often introduces systematic, accent-specific errors that filtering fails to fix. We ask: How can we correct these recurring biases without target ground truth? We propose a simple parameter-space correction: in a source domain containing both real and pseudo-labeled data, two ASR models are fine-tuned from the same initialization, one on ground-truth labels and the other on pseudo-labels, and their weight difference forms a correction vector that captures pseudo-label biases. When applied to a pseudo-labeled target model, this vector enhances recognition, achieving up to a 35% relative Word Error Rate (WER) reduction on AfriSpeech-200 across ten African accents with the Whisper tiny model. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_08047 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Pseudo2Real: Task Arithmetic for Pseudo-Label Correction in Automatic Speech Recognition Lin, Yi-Cheng Liang, Yu-Hsuan Li Su, Hsuan Lin, Tzu-Quan Chen, Shang-Tse Chen, Yun-Nung Lee, Hung-yi Audio and Speech Processing Computation and Language Robust ASR under domain shift is crucial because real-world systems encounter unseen accents and domains with limited labeled data. Although pseudo-labeling offers a practical workaround, it often introduces systematic, accent-specific errors that filtering fails to fix. We ask: How can we correct these recurring biases without target ground truth? We propose a simple parameter-space correction: in a source domain containing both real and pseudo-labeled data, two ASR models are fine-tuned from the same initialization, one on ground-truth labels and the other on pseudo-labels, and their weight difference forms a correction vector that captures pseudo-label biases. When applied to a pseudo-labeled target model, this vector enhances recognition, achieving up to a 35% relative Word Error Rate (WER) reduction on AfriSpeech-200 across ten African accents with the Whisper tiny model. |
| title | Pseudo2Real: Task Arithmetic for Pseudo-Label Correction in Automatic Speech Recognition |
| topic | Audio and Speech Processing Computation and Language |
| url | https://arxiv.org/abs/2510.08047 |