You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

You agree to the terms of the LICENSE when using this dataset. Please read it carefully.
このデータセットを使用する際には、LICENSEの条件に同意したことになります。必ずお読みください。

Log in or Sign Up to review the conditions and access this dataset content.

😁 Ehehe Corpus 🤣

Audio dataset of laughter performed by Japanese voice actors

日本語はこちら

このデータセットは、著作権法第三十条の四の情報解析(機械学習等)の目的でのみ使用が許可されています。それ以外の用途での使用はライセンスにより禁止されています。

This dataset is only permitted for use under Article 30-4 of the Copyright Law of Japan for data analysis (such as machine learning) purposes. Any use for purposes other than those specified is prohibited by the license.

Dataset Summary

  • A high-quality dataset of all only laughter voice acting recordings by Japanese professional voice actors, recorded in a studio, free of noise and background music (both male and female characters).
  • All the audio files are manually checked to ensure that they are laughter voices.
  • Details:
    • Duration: About 5.13 hours
    • Audio format: 44.1kHz 16bit mono WAV
    • Voice actors: 350 (+ mob) (but some voice actors may have multiple identifiers)
    • Audio files: 16,415 files
    • File size: About 1.51GB
    • Each audio file contains only one voice actor's performance (no audio files with multiple voice actors)
  • This dataset also includes unique expressions of laughter such as 'wa-hahaha', 'gahaha', 'nishishi', and 'niyaniya' etc, which are commonly heard in Japanese fictional characters.
  • The dataset is organized into folders for each voice actors (represented by uuid.uuid4().hex[:8] random alphanumeric strings), and in each folder, laughter audio files are divided into characters played by the voice actor.
  • The transcription of each audio file (by faster-whisper medium) is also included.

stats.png

  • The code used to create this dataset is open-source, so please create and publish new datasets using it!

Supported Tasks and Leaderboards

This dataset may be useful for research and development of laughter-related tasks such as laughter synthesis and laughter detection, especially in the context of Japanese otaku culture. See Curation Rationale for more details.

Languages

All audio in the dataset is in Japanese. Some laughter expressions are unique to Japanese fictional characters.

Dataset Structure

├── metadata.csv
├── original_transcriptions.csv
├── stats.csv
└── data/
      ├── {uuid1}/
      │     ├── 1/
      │     │   ├── {uuid_a}.wav
      │     │   ├── {uuid_b}.wav
      │     │   ├── ...
      │     ├── 2/
      │     │   ├── {uuid_c}.wav
      │     │   ├── {uuid_d}.wav
      │     │   ├── ...
      │     ├── ...
      ├── {uuid2}/
      │     ├── 1/
      |    ...
      └── mob/
            └── 1/
                  ├── {uuid_e}.wav
                  ├── {uuid_f}.wav
                  ├── ...

Here,

  • {uuid1}, {uuid2}: Random 8-character alphanumeric string representing a voice actor.
  • 1, 2, ... in each voice actor folder: Different characters played by the voice actor.
  • mob: Folder for recordings of mob characters (characters with no specific name or role) and unknown voice actors.

stats.csv:

actor,num_characters,total_duration_sec,f0_mean
mob,1,667.22,193.06
3b555351,11,629.96,177.78
3dce70d9,10,589.17,238.51
c1ab42e2,2,537.81,302.75
...
  • actor: Voice actor identifier (8-character random alphanumeric string)
  • num_characters: Number of characters played by the voice actor
  • total_duration_sec: Total duration of audio files (seconds)
  • f0_mean: Mean of fundamental frequency (Hz)
  • The order is sorted by total duration in descending order.

metadata.csv:

file_name,transcription
data/001ed88f/1/00e60b8a.wav,うふふふっ
data/001ed88f/1/ea143be4.wav,あははは
data/006635cc/1/1111f11f.wav,ふふふっ
...

Note: The transcription is done by the faster-whisper medium model (original_transcriptions.csv) and then normalized using normalize.py. It is not perfect and may contain errors or hallucinations or empty transcriptions.

TODO: Manual correction of transcription.

(Whisper's large models or above tend to have more hallucinations in laughter voices.)

Download

Using the huggingface-cli can be convenient. Create a token from the Hugging Face settings page and then log in using the following commands:

pip install -U "huggingface_hub[cli]"
huggingface-cli login

To download all:

huggingface-cli download litagin/ehehe-corpus --repo-type dataset --local-dir path/to/download/

For more details, refer to the Hugging Face CLI documentation.

Or git clone (Hugging Face token required):

git lfs install
git clone https://<user_name>:<token>@huggingface.co/datasets/litagin/ehehe-corpus

Dataset Creation

Curation Rationale

In communication, voice plays a role not only in conveying information but also in transmitting emotions and intentions. Therefore, non-verbal vocal sounds, in addition to normal spoken language, are significant. Among these non-verbal sounds, laughter is one of the most frequently used in everyday life, and studies on laughter (like laughter detection and laughter synthesis) have been extensively conducted.

However, existing research on laughter primarily focuses on "natural laughter that occurs in daily life," with little attention given to "laughter expressions in fiction such as anime and comics." Non-verbal vocal sounds are an important aspect of culture, and in Japanese otaku culture, particularly, they play a crucial role as part of a character's personality and expression technique.

For studying laughter in fiction, an actual dataset of voice actor performances is essential, yet to my knowledge, no such dataset of acted laughter exists. This dataset, when compared to existing laughter corpora, would be distinctive in the following points:

  • High-quality Japanese laughter recordings by professional voice actors.
  • Includes types of laughter unique to fictional Japanese characters, which are not found in general conversation among Japanese people and thus cannot be collected from sources like YouTube. These include:
    • Unique laughter expressions (e.g., "wa-ha-ha", "ga-ha-ha", "ku-hi-hi", "nishishi-").
    • Laughter for evasion, embarrassed laughter, maniacal laughter, exaggeratedly acted laughter, among various others.

With these features, the corpus is expected to provide a better understanding of the vocal aspects of Japanese otaku culture and contribute to the research and development of laughter detection (like detecting characters' laughter in anime) and laughter synthesis technologies for fiction.

Source Data

Recordings from PC games that were purchased through legal means and are personally owned.

Initial Data Collection and Normalization

The dataset was collected by the following process:

  1. Cut out each voice by -40dB threshold (using pydub)
  2. Transcribe the text of each voice (using Whisper Medium model)
  3. Detect laughter by its transcription (using a simple regex pattern)
  4. At this point, the numbers of audio files were 46,554, but contains many non-laughter voices. So I manually checked all the audio files and removed non-laughter voices.
  5. After manual checking, some audio files were replaced with the original audio (since often a single laughter audio file was divided into multiple files), and also some files were manually cut from the original audio files for better cutting points. (I'm sorry that I haven't done this for all the files)

Annotations

  • Fundamental frequency of audio files was calculated using FCPE.
  • All the audio files were transcribed using the faster-whisper medium model.

Personal and Sensitive Information

To prevent misuse for enjoyment purposes, the following measures have been taken:

  • Game names and character names are concealed, no categorization by game is done in folder organization, and random alphanumeric strings are used as voice actor identifiers.
  • All the audio files are also random alphanumeric strings.

Considerations for Using the Data

Discussion of Biases

Due to its nature, the dataset may exhibit certain biases, such as:

  • A tendency for a larger volume of data for female characters.

Other Known Limitations

  • Since all laughter voices were manually checked, there may be non-laughter voices mixed in due to mistakes.
  • Trimming of audio is done automatically, so there may be unnatural breaks or starts in the audio.
  • There may be non-laughter voices at the beginning or end of the audio.
  • The same voice actor may have multiple identifiers and may have audios in mob directory.
  • Some files include:
    • Audio processed to sound like it has an echo.
    • Audio processed to sound as if it's coming through a phone or from behind a wall.

Additional Information

Licensing Information

Please refer to LICENSE for details. It is essential to read and understand the LICENSE before using this dataset to ensure compliance with its terms.

Disclaimer

  • The providers of this dataset are not responsible for any troubles or damages arising from the use of this dataset.
  • Users must comply with the laws of their country or region when using this dataset.

The legal basis for publishing this dataset is as follows: Copyright Law of Japan (Law No. 48 of May 6, 1970) Article 30-4:

(Quotation starts)

Article 30-4: It is permissible to exploit a work, in any way and to the extent considered necessary, in any of the following cases, or in any other case in which it is not a person's purpose to personally enjoy or cause another person to enjoy the thoughts or sentiments expressed in that work; provided, however, that this does not apply if the action would unreasonably prejudice the interests of the copyright owner in light of the nature or purpose of the work or the circumstances of its exploitation:

(i) if it is done for use in testing to develop or put into practical use technology that is connected with the recording of sounds or visuals of a work or other such exploitation;

(ii) if it is done for use in data analysis (meaning the extraction, comparison, classification, or other statistical analysis of the constituent language, sounds, images, or other elemental data from a large number of works or a large volume of other such data; the same applies in Article 47-5, paragraph (1), item (ii));

(iii) if it is exploited in the course of computer data processing or otherwise exploited in a way that does not involve what is expressed in the work being perceived by the human senses (for works of computer programming, such exploitation excludes the execution of the work on a computer), beyond as set forth in the preceding two items.

(End of quotation)

  • This dataset is considered to fall under the second category mentioned above.
  • The dataset is structured to meet the condition of "person's purpose to personally enjoy or cause another person to enjoy the thoughts or sentiments expressed in that work", as specified in LICENSE, and users are prohibited from using it for enjoyment purposes.
  • Regarding "this does not apply if the action would unreasonably prejudice the interests of the copyright owner in light of the nature or purpose of the work or the circumstances of its exploitation", this dataset conceals the source game, voice actor names, and character names, and the order of the audio files is randomized. Additionally, identifiers with the same source are not disclosed, making it impossible to use this dataset for the original purpose of the work (enjoying the game's scenario with voice and images), and thus it is believed that the publication of this dataset does not unfairly harm the interests of the copyright holder.
  • Even if a user tries to ignore the license and use it for enjoyment, all the audio files are laughter voices, so a user cannot enjoy the original work's scenario or story.
  • As stated in LICENSE, any use that "unfairly harms the interests of the copyright holder", as well as actions that hinder the considerations mentioned above (providing information to third parties about the voice actors or original works, or redistributing in a way that makes these associations identifiable), are prohibited.

😁 えへへコーパス 🤣 (日本語版README)

日本人プロ声優による笑い声演技音声データセット

English version here

Dataset Description

Dataset Summary

  • 日本人プロ声優による高音質(スタジオ録音)でノイズ・BGM等無しのキャラクター笑い声のみからなる演技音声データセット(男性・女性キャラクター両方含む)
  • 全ての音声ファイルは笑い声であることを手動チェックにより確認済み
  • 詳細:
    • 時間長: 約5.13時間
    • オーディオ形式: 44.1kHz 16bit モノラル WAV
    • 声優数: 350 (+ mob) (ただし一部声優に複数の識別子がある可能性あり)
    • ファイル数: 16,415 ファイル
    • ファイルサイズ: 約1.51GB
    • 各音声は一人の声優の演技のみを含む(複数の声優による笑い声音声は含まれない)
  • このデータセットには、日本の萌え文化特有の「わーはっはっは」・「がはは」・「にしし」・「にやにや」等の特殊な笑い声も含まれている
  • データセットは各声優ごとにフォルダ分けされており(uuid.uuid4().hex[:8]のランダムな英数字文字列で表現)、各フォルダ内には演じたキャラクターごとに笑い声音声ファイルが分かれて配置されている
  • 各音声ファイルの書き起こし(faster-whisper mediumによる)も含まれている

stats.png

  • データセット作成に使用したコードを公開しているので、ぜひ新しいデータセットを作成して公開してください!

Supported Tasks and Leaderboards

日本のアニメ等のオタク文化に特化した笑い声合成や笑い声検出等の研究開発に役立つことが期待される。詳しくは下のCuration Rationaleを参照。

Languages

データセット内の全ての音声は日本語。一部の笑い声表現は日本の架空のキャラクターに特有のもの。

Dataset Structure

├── metadata.csv
├── original_transcriptions.csv
├── stats.csv
└── data/
      ├── {uuid1}/
      │     ├── 1/
      │     │   ├── {uuid_a}.wav
      │     │   ├── {uuid_b}.wav
      │     │   ├── ...
      │     ├── 2/
      │     │   ├── {uuid_c}.wav
      │     │   ├── {uuid_d}.wav
      │     │   ├── ...
      │     ├── ...
      ├── {uuid2}/
      │     ├── 1/
      |    ...
      └── mob/
            └── 1/
                  ├── {uuid_e}.wav
                  ├── {uuid_f}.wav
                  ├── ...

ここで、

  • {uuid1}, {uuid2}: 声優を表すランダムな8文字の英数字文字列
  • 各声優フォルダ内の1, 2, ...: 声優が演じた異なるキャラクター
  • mob: モブキャラクター(特定の名前や役割のないキャラクター)や不明な声優の音声のフォルダ

stats.csv:

actor,num_characters,total_duration_sec,f0_mean
mob,1,667.22,193.06
3b555351,11,629.96,177.78
3dce70d9,10,589.17,238.51
c1ab42e2,2,537.81,302.75
...
  • actor: 声優識別子(8文字のランダムな英数字文字列)
  • num_characters: 声優が演じたキャラクター数
  • total_duration_sec: 音声ファイルの合計時間(秒)
  • f0_mean: 基本周波数の平均値(Hz)
  • 合計時間で降順に並べられている

metadata.csv:

file_name,transcription
data/001ed88f/1/00e60b8a.wav,うふふふっ
data/001ed88f/1/ea143be4.wav,あははは
data/006635cc/1/1111f11f.wav,ふふふっ
...

注意: 書き起こしはfaster-whisper mediumモデルによって行われ(original_transcriptions.csv)、その後でnormalize.pyで正規化されている。完璧ではなく、誤りやハルシネーション、空の書き起こしを含む可能性がある。

(Whisperのlargeモデル以上だと笑い声に関してよりハルシネーションがひどくなる傾向が高いようである。)

Download

huggingface-cliを使うと便利です。 Huggung Faceの設定ページからトークンを作り、以下でログインします。

pip install -U "huggingface_hub[cli]"
huggingface-cli login

全てダウンロードするには:

huggingface-cli download litagin/ehehe-corpus --repo-type dataset --local-dir path/to/download/

詳細はHugging Face CLI documentationを参照してください。

またはgit clone (Hugging Faceのトークンが必要です):

git lfs install
git clone https://<user_name>:<token>@huggingface.co/datasets/litagin/ehehe-corpus

Dataset Creation

Curation Rationale

音声はコミュニケーションにおいて、単に情報を伝えるだけでなく、感情や意図を伝える役割を果たし、よって言語を読み上げる通常の発話だけではなく、非言語音声も重要な役割を果たしている。非言語音声の中でも笑い声は日常で最も使われ、また笑い声に関する研究(笑い声検出や笑い声合成)も多く試みられている。

しかし、既存の笑い声に関する研究は「日常で現れる自然な笑い声」に焦点を当てており、「アニメや漫画等のフィクションにおける笑い声表現」に関する研究はほとんど見受けられない。非言語音声は文化の大切な要素であり、特に日本文化においては、キャラクターの個性や演出の一部として重要な役割を果たしている。

このようなフィクションにおける笑い声を研究する際には、実際の声優による演技音声データセットが必要であるが、既存の笑い声コーパスはそもそも数が限られており、その上で演技された笑い声データセットは私の知る限り存在しない。 具体的には、これまでの存在する笑い声コーパスと比べて、以下の点が特徴的である:

  • プロ声優による高音質の日本語の笑い声録音
  • 日本の架空のキャラクターに特有の以下の種類の笑い声が含まれており、日本人による一般的な会話では聞かれないため、YouTube等で収集することができない:
    • 特殊な笑い声表現(例:わーはっはっは、がはは、くひひ、にっしし~等)
    • ごまかす笑い声、恥ずかしがる笑い声、狂気の笑い声、大げさ演技した笑い声等、様々な笑い声

これらの特徴を持つことで、このコーパスは、日本のオタク文化の音声的側面によりよい理解を提供し、また特徴を活かした笑い声検出(アニメ等でのキャラクターの笑い声の検出)や、フィクションのような笑い声合成技術の研究開発に役立つことが期待される。

Source Data

正規の手段で購入して個人的に所持しているPCゲームから録音したもの

Initial Data Collection and Normalization

データセットは以下の手順で収集された。GitHubでLaughter Collectorとして収集するのに使用したスクリプト群を公開しているので、興味があればそちらを参照してください。

  1. 各声を-40dBの閾値でカットアウト(pydubを使用)
  2. 各声のテキストを書き起こし(Whisper Mediumモデルを使用)
  3. 書き起こしによって笑い声を検出(簡単な正規表現パターンを使用)
  4. この時点で音声ファイルの数は46,554だったが、多くの非笑い声が含まれていたため、手動で全ての音声ファイルをチェックし、非笑い声を削除した
  5. 手動チェック後、一部の音声ファイルは元の音声で置き換えた(1つの笑い声音声ファイルが複数のファイルに分割されていることがあるため)。また、一部のファイルは途中で途切れていたので、元の音声ファイルから手動でカットし直した(ただし全てのファイルに対してはこれを行っていないことをお詫び申し上げます)

Annotations

  • 音声ファイルの基本周波数はFCPEを使用して計算した
  • 音声ファイルの書き起こしは全てWhisper Mediumモデルを使用して行った

Personal and Sensitive Information

享受目的での利用を防ぐため、以下のような手段を取っている。

  • ゲーム名やキャラクター名を伏せ、ゲームによるフォルダ分け類別はせず、また声優識別子にはランダムな英数字文字列を使用
  • 全ての音声ファイルもランダムな英数字文字列で表現

Considerations for Using the Data

Discussion of Biases

性質上、以下のようなバイアスがある可能性がある。

  • 女性キャラクターの方がデータ量が多い傾向がある

Other Known Limitations

  • 笑い声かどうかを全て手動でチェックしたため、ミスにより笑い声には聞こえない音声も混入している可能性がある。
  • 音声のトリミングは機械的に行われているので、不自然な箇所で途切れている・途中から始まっている音声がある可能性がある。
  • 冒頭や末尾等に笑い声以外の音声が入っている可能性がある。
  • 同一声優が複数の識別子を持ったり、mobディレクトリにその声優の音声がある可能性がある。
  • 以下のような音声が少数入っている。
    • エコーが入ったような加工がされた音声
    • 電話越し・壁越しであるかのような加工がされた音声

Additional Information

Licensing Information

LICENSEを参照。このデータセットを利用する場合は、必ずLICENSEを読んで利用条件を確認すること。

Disclaimer

  • このデータセットの利用によって発生したいかなるトラブルや損害に対しても、データセットの提供者は責任を負わない。
  • このデータセットの利用に際して、自身の国または地域の法律に従うこと。

このデータセットを公開している根拠は、以下の通り。 著作権法(昭和45年5月6日法律第48号)第三十条の四:

(以下引用)

著作物は、次に掲げる場合その他の当該著作物に表現された思想又は感情を自ら享受し又は他人に享受させることを目的としない場合には、その必要と認められる限度において、いずれの方法によるかを問わず、利用することができる。ただし、当該著作物の種類及び用途並びに当該利用の態様に照らし著作権者の利益を不当に害することとなる場合は、この限りでない。

 一 著作物の録音、録画その他の利用に係る技術の開発又は実用化のための試験の用に供する場合

 二 情報解析(多数の著作物その他の大量の情報から、当該情報を構成する言語、音、影像その他の要素に係る情報を抽出し、比較、分類その他の解析を行うことをいう。第四十七条の五第一項第二号において同じ。)の用に供する場合

 三 前二号に掲げる場合のほか、著作物の表現についての人の知覚による認識を伴うことなく当該著作物を電子計算機による情報処理の過程における利用その他の利用(プログラムの著作物にあつては、当該著作物の電子計算機における実行を除く。)に供する場合

(引用終わり)

  • このデータセットは、上記の第二号に該当すると考えられる。
  • 「著作物に表現された思想又は感情を自ら享受し又は他人に享受させることを目的としない場合」という条件を満たすように配慮したデータセットの構造となっており、LICENSEにある通り、利用者は享受目的の利用を禁止されている。
  • 「当該著作物の種類及び用途並びに当該利用の態様に照らし著作権者の利益を不当に害することとなる場合」については、本データセットでは参照元や声優名やキャラクター名を伏せている上に音声の順番もシャッフルされており、また同一参照元を持つ識別子も公開していないことから、当該著作物(ゲーム)の使用用途(シナリオを音声と絵をあわせて楽しむ)で利用することは不可能であり、このデータセットの公開によって著作権者の利益を不当に害することはないと考えられる。
  • もし仮に利用者がライセンスを無視し享受利用しようとしたとしても、データセットには笑い声しか含まれていないため、そこから元のゲームのシナリオを再現し享受することは完全に不可能である。
  • LICENSEにある通り、「当該著作物の種類及び用途並びに当該利用の態様に照らし著作権者の利益を不当に害することとなる」ような利用方法は禁じられており、また以上の配慮を妨げるような行為(声優や元著作物との対応についての第三者への情報提供やそれらが分かるような形での再配布)は禁止されている。
Downloads last month
0
Edit dataset card