Conference paper Open Access

On Using SpecAugment for End-to-End Speech Translation

Bahar, Parnia; Zeyer, Albert; Schlüter, Ralf; Ney, Hermann


JSON Export

{
  "files": [
    {
      "links": {
        "self": "https://zenodo.org/api/files/b5358cd0-864a-4253-8e1d-698e78fbd1ed/IWSLT2019_paper_19.pdf"
      }, 
      "checksum": "md5:b101a778188679cfbdd072bdb714062c", 
      "bucket": "b5358cd0-864a-4253-8e1d-698e78fbd1ed", 
      "key": "IWSLT2019_paper_19.pdf", 
      "type": "pdf", 
      "size": 768990
    }
  ], 
  "owners": [
    50447
  ], 
  "doi": "10.5281/zenodo.3525010", 
  "stats": {
    "version_unique_downloads": 66.0, 
    "unique_views": 83.0, 
    "views": 94.0, 
    "version_views": 96.0, 
    "unique_downloads": 66.0, 
    "version_unique_views": 85.0, 
    "volume": 58443240.0, 
    "version_downloads": 76.0, 
    "downloads": 76.0, 
    "version_volume": 58443240.0
  }, 
  "links": {
    "doi": "https://doi.org/10.5281/zenodo.3525010", 
    "conceptdoi": "https://doi.org/10.5281/zenodo.3525009", 
    "bucket": "https://zenodo.org/api/files/b5358cd0-864a-4253-8e1d-698e78fbd1ed", 
    "conceptbadge": "https://zenodo.org/badge/doi/10.5281/zenodo.3525009.svg", 
    "html": "https://zenodo.org/record/3525010", 
    "latest_html": "https://zenodo.org/record/3525010", 
    "badge": "https://zenodo.org/badge/doi/10.5281/zenodo.3525010.svg", 
    "latest": "https://zenodo.org/api/records/3525010"
  }, 
  "conceptdoi": "10.5281/zenodo.3525009", 
  "created": "2019-11-01T12:59:15.965475+00:00", 
  "updated": "2020-01-20T16:44:30.391967+00:00", 
  "conceptrecid": "3525009", 
  "revision": 3, 
  "id": 3525010, 
  "metadata": {
    "access_right_category": "success", 
    "doi": "10.5281/zenodo.3525010", 
    "description": "<p>This work investigates a simple data augmentation technique, SpecAugment, for end-to-end speech translation. SpecAugment is a low-cost implementation method applied directly to the audio input features and it consists of masking blocks of frequency channels, and/or time steps. We apply SpecAugment on end-to-end speech translation tasks and achieve up to +2.2% BLEU&nbsp;on LibriSpeech Audiobooks En&rarr;Fr and +1.2% on IWSLT TED-talks En&rarr;De by alleviating overfitting to some extent. We also examine the effectiveness of the method in a variety of data scenarios and show that the method also leads to significant improvements in various data conditions irrespective of the amount of training data.</p>", 
    "license": {
      "id": "CC-BY-4.0"
    }, 
    "title": "On Using SpecAugment for End-to-End Speech Translation", 
    "relations": {
      "version": [
        {
          "count": 1, 
          "index": 0, 
          "parent": {
            "pid_type": "recid", 
            "pid_value": "3525009"
          }, 
          "is_last": true, 
          "last_child": {
            "pid_type": "recid", 
            "pid_value": "3525010"
          }
        }
      ]
    }, 
    "communities": [
      {
        "id": "iwslt2019"
      }
    ], 
    "publication_date": "2019-11-02", 
    "creators": [
      {
        "affiliation": "Human Language Technology and Pattern Recognition Group Computer Science Department, RWTH Aachen University, 52062 Aachen, Germany &  AppTek, 52062 Aachen, Germany", 
        "name": "Bahar, Parnia"
      }, 
      {
        "affiliation": "Human Language Technology and Pattern Recognition Group Computer Science Department, RWTH Aachen University, 52062 Aachen, Germany &  AppTek, 52062 Aachen, Germany", 
        "name": "Zeyer, Albert"
      }, 
      {
        "affiliation": "Human Language Technology and Pattern Recognition Group, Computer Science Department, RWTH Aachen University, 52062 Aachen, Germany", 
        "name": "Schl\u00fcter, Ralf"
      }, 
      {
        "affiliation": "Human Language Technology and Pattern Recognition Group Computer Science Department, RWTH Aachen University, 52062 Aachen, Germany &  AppTek, 52062 Aachen, Germany", 
        "name": "Ney, Hermann"
      }
    ], 
    "access_right": "open", 
    "resource_type": {
      "subtype": "conferencepaper", 
      "type": "publication", 
      "title": "Conference paper"
    }, 
    "related_identifiers": [
      {
        "scheme": "doi", 
        "identifier": "10.5281/zenodo.3525009", 
        "relation": "isVersionOf"
      }
    ]
  }
}
96
76
views
downloads
All versions This version
Views 9694
Downloads 7676
Data volume 58.4 MB58.4 MB
Unique views 8583
Unique downloads 6666

Share

Cite as