| 電子情報通信学会 研究会発表申込システム 講演論文 詳細 |
技報閲覧サービス [ログイン] 技報アーカイブ |
| トップに戻る | 前のページに戻る | [Japanese] / [English] |
| 講演抄録／キーワード | ||
| 講演名 | 2023-03-01 13:45 ［招待講演］Speech and Language Research in the Google Tokyo Office ○Michiel Bacchiani（Google）  EA2022-116 SIP2022-160 SP2022-80 |
|
| 抄録 | （和） | This talk will consist of three parts. In the first part of the talk, I will reflect on some lessons learned from the activities that earned me the 2022 elevation to IEEE Fellow with the citation “for leadership in commercial automatic speech recognition systems”. In the second part of the talk, I will present our recent work on “Helpful Neighbors”. Predicting the reading a Japanese place name based on the orthography alone is notoriously difficult, even for native speakers. This project proposes a novel machine learning model to predict the reading of place names in Japan using reading information from neighboring locations as contextual input. In the third part of this talk, I will present our recent work on Wavefit, a neural vocoders based on denoising diffusion probabilistic models (DDPM). This model enhances previously proposed DDPMs with an adaptive diffusion noise and incorporates the essence of Generative Adversarial Networks. The speech synthesized by this algorithm is in terms of quality on par with human speech and the time to synthesize speech is 240 times faster than the WaveRNN algorithm used in many of our products today. |
| （英） | This talk will consist of three parts. In the first part of the talk, I will reflect on some lessons learned from the activities that earned me the 2022 elevation to IEEE Fellow with the citation “for leadership in commercial automatic speech recognition systems”. In the second part of the talk, I will present our recent work on “Helpful Neighbors”. Predicting the reading a Japanese place name based on the orthography alone is notoriously difficult, even for native speakers. This project proposes a novel machine learning model to predict the reading of place names in Japan using reading information from neighboring locations as contextual input. In the third part of this talk, I will present our recent work on Wavefit, a neural vocoders based on denoising diffusion probabilistic models (DDPM). This model enhances previously proposed DDPMs with an adaptive diffusion noise and incorporates the essence of Generative Adversarial Networks. The speech synthesized by this algorithm is in terms of quality on par with human speech and the time to synthesize speech is 240 times faster than the WaveRNN algorithm used in many of our products today. | |
| キーワード | （和） | Letter-to-sound / denoising diffusion probabilistic models / Text to speech / Speech and Language Processing / / / / |
| （英） | Letter-to-sound / denoising diffusion probabilistic models / Text to speech / Speech and Language Processing / / / / | |
| 文献情報 | 信学技報, vol. 122, no. 389, SP2022-80, pp. 239-240, 2023年2月. | |
| 資料番号 | SP2022-80 | |
| 発行日 | 2023-02-21 (EA, SIP, SP) | |
| ISSN | Online edition: ISSN 2432-6380 | |
| 著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します．(許諾番号：10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) | |
| PDFダウンロード | EA2022-116 SIP2022-160 SP2022-80 |
|
| 研究会情報 | |
| 研究会 | SP IPSJ-SLP EA SIP |
| 開催期間 | 2023-02-28 - 2023-03-01 |
| 開催地（和） | 沖縄県立博物館・美術館 |
| 開催地（英） | |
| テーマ（和） | 音声，応用／電気音響, 信号処理，一般 |
| テーマ（英） | |
| 講演論文情報の詳細 | |
| 申込み研究会 | SP |
| 会議コード | 2023-02-SP-SLP-EA-SIP |
| 本文の言語 | 英語 |
| タイトル（和） | |
| サブタイトル（和） | |
| タイトル（英） | Speech and Language Research in the Google Tokyo Office |
| サブタイトル（英） | |
| キーワード(1)（和/英） | Letter-to-sound / Letter-to-sound |
| キーワード(2)（和/英） | denoising diffusion probabilistic models / denoising diffusion probabilistic models |
| キーワード(3)（和/英） | Text to speech / Text to speech |
| キーワード(4)（和/英） | Speech and Language Processing / Speech and Language Processing |
| キーワード(5)（和/英） | / |
| キーワード(6)（和/英） | / |
| キーワード(7)（和/英） | / |
| キーワード(8)（和/英） | / |
| 第1著者 氏名（和/英/ヨミ） | Michiel Bacchiani / Michiel Bacchiani / |
| 第1著者 所属（和/英） | Google (略称： Google) Google (略称： Google) |
| 講演者 | 第1著者 |
| 発表日時 | 2023-03-01 13:45:00 |
| 発表時間 | 45分 |
| 申込先研究会 | SP |
| 資料番号 | EA2022-116, SIP2022-160, SP2022-80 |
| 巻番号（vol） | vol.122 |
| 号番号（no） | no.387(EA), no.388(SIP), no.389(SP) |
| ページ範囲 | pp.239-240 |
| ページ数 | 2 |
| 発行日 | 2023-02-21 (EA, SIP, SP) |
[研究会発表申込システムのトップページに戻る]