How do Hyenas deal with Human Speech? Speech Recognition and Translation with ConfHyena
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2024
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866914687012044800 |
|---|---|
| author | Gaido, Marco Papi, Sara Negri, Matteo Bentivogli, Luisa |
| author_facet | Gaido, Marco Papi, Sara Negri, Matteo Bentivogli, Luisa |
| contents | The attention mechanism, a cornerstone of state-of-the-art neural models, faces computational hurdles in processing long sequences due to its quadratic complexity. Consequently, research efforts in the last few years focused on finding more efficient alternatives. Among them, Hyena (Poli et al., 2023) stands out for achieving competitive results in both language modeling and image classification, while offering sub-quadratic memory and computational complexity. Building on these promising results, we propose ConfHyena, a Conformer whose encoder self-attentions are replaced with an adaptation of Hyena for speech processing, where the long input sequences cause high computational costs. Through experiments in automatic speech recognition (for English) and translation (from English into 8 target languages), we show that our best ConfHyena model significantly reduces the training time by 27%, at the cost of minimal quality degradation (~1%), which, in most cases, is not statistically significant. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2402_13208 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | How do Hyenas deal with Human Speech? Speech Recognition and Translation with ConfHyena Gaido, Marco Papi, Sara Negri, Matteo Bentivogli, Luisa Computation and Language Artificial Intelligence The attention mechanism, a cornerstone of state-of-the-art neural models, faces computational hurdles in processing long sequences due to its quadratic complexity. Consequently, research efforts in the last few years focused on finding more efficient alternatives. Among them, Hyena (Poli et al., 2023) stands out for achieving competitive results in both language modeling and image classification, while offering sub-quadratic memory and computational complexity. Building on these promising results, we propose ConfHyena, a Conformer whose encoder self-attentions are replaced with an adaptation of Hyena for speech processing, where the long input sequences cause high computational costs. Through experiments in automatic speech recognition (for English) and translation (from English into 8 target languages), we show that our best ConfHyena model significantly reduces the training time by 27%, at the cost of minimal quality degradation (~1%), which, in most cases, is not statistically significant. |
| title | How do Hyenas deal with Human Speech? Speech Recognition and Translation with ConfHyena |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2402.13208 |