Assessing the Impact of Anisotropy in Neural Representations of Speech: A Case Study on Keyword Spotting
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866916792464572416 |
|---|---|
| author | Wisniewski, Guillaume Guillaume, Séverine Fernández, Clara Rosina |
| author_facet | Wisniewski, Guillaume Guillaume, Séverine Fernández, Clara Rosina |
| contents | Pretrained speech representations like wav2vec2 and HuBERT exhibit strong anisotropy, leading to high similarity between random embeddings. While widely observed, the impact of this property on downstream tasks remains unclear. This work evaluates anisotropy in keyword spotting for computational documentary linguistics. Using Dynamic Time Warping, we show that despite anisotropy, wav2vec2 similarity measures effectively identify words without transcription. Our results highlight the robustness of these representations, which capture phonetic structures and generalize across speakers. Our results underscore the importance of pretraining in learning rich and invariant speech representations. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2506_11096 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Assessing the Impact of Anisotropy in Neural Representations of Speech: A Case Study on Keyword Spotting Wisniewski, Guillaume Guillaume, Séverine Fernández, Clara Rosina Sound Artificial Intelligence Computation and Language Audio and Speech Processing Pretrained speech representations like wav2vec2 and HuBERT exhibit strong anisotropy, leading to high similarity between random embeddings. While widely observed, the impact of this property on downstream tasks remains unclear. This work evaluates anisotropy in keyword spotting for computational documentary linguistics. Using Dynamic Time Warping, we show that despite anisotropy, wav2vec2 similarity measures effectively identify words without transcription. Our results highlight the robustness of these representations, which capture phonetic structures and generalize across speakers. Our results underscore the importance of pretraining in learning rich and invariant speech representations. |
| title | Assessing the Impact of Anisotropy in Neural Representations of Speech: A Case Study on Keyword Spotting |
| topic | Sound Artificial Intelligence Computation and Language Audio and Speech Processing |
| url | https://arxiv.org/abs/2506.11096 |