PodCastle の提案: 音声認識研究2.0を目指して  [in Japanese] PodCastle: A Proposal for Speech Recognition Research 2.0  [in Japanese]

Access this Article

Search this Article

Author(s)

Abstract

本稿では、Web 2.0 に基づく Webサービスを提供し、音声認識性能の現状を積極的に開示することで、不特定多数のユーザの協力を得て音声認識技術を発展させていく研究アプローチ「音声認識研究 2.0」を提案する。我々は、これを具現化した音声認識のキラーアプリケーションを目指して、音声認識に基づくポッドキャスト検索サービス PodCastle の公開を開始した。PodCastle では、ユーザがWeb上の日本語のポッドキャストを全文検索できるだけでなく、認識結果の全文テキストも閲覧でき、さらに誤認識箇所を容易に訂正することもできる。これにより、ユーザが利用しながら訂正すると認識性能と検索性能が向上し、さらなる利用が促せるというポジティブスパイラルが生じることが期待できる。In this paper, we propose "Speech Recognition Research 2.0" a research approach that provides users with a web service based on Web 2.0 to disclose state-of-the-art speech recognition performances and to promote speech recognition technologies in cooperation with anonymous users. In the quest for a killer application of speech recognition which embodies this approach, we launched a public web service "PodCastle" for searching podcasts on the basis of speech recognition. PodCastle enables users to accomplish a full-text search of Japanese podcasts on the web, read full texts of their speech recognition results, and easily correct recognition errors. We can thus expect a positive spiral where the improvement of the recognition and search performances through the correction by users encourages further usage of the web service.

In this paper, we propose "Speech Recognition Research 2.0" a research approach that provides users with a web service based on Web 2.0 to disclose state-of-the-art speech recognition performances and to promote speech recognition technologies in cooperation with anonymous users. In the quest for a killer application of speech recognition which embodies this approach, we launched a public web service "PodCastle" for searching podcasts on the basis of speech recognition. PodCastle enables users to accomplish a full-text search of Japanese podcasts on the web, read full texts of their speech recognition results, and easily correct recognition errors. We can thus expect a positive spiral where the improvement of the recognition and search performances through the correction by users encourages further usage of the web service.

Journal

  • IPSJ SIG Notes

    IPSJ SIG Notes 2007(11(2007-SLP-065)), 35-40, 2007-02-09

    Information Processing Society of Japan (IPSJ)

Cited by:  8

Codes

  • NII Article ID (NAID)
    110006224483
  • NII NACSIS-CAT ID (NCID)
    AN10442647
  • Text Lang
    JPN
  • Article Type
    Journal Article
  • Data Source
    CJPref  NII-ELS  IPSJ 
Page Top