README.md

---
language:
- ja
license: gpl
size_categories:
- 100K<n<1M
task_categories:
- automatic-speech-recognition
pretty_name: Japanese-Anime-Speech-V2
dataset_info:
  features:
  - name: audio
    dtype:
      audio:
        sampling_rate: 16000
  - name: transcription
    dtype: string
  splits:
  - name: train
    num_bytes: 21289144155.152
    num_examples: 300506
  download_size: 24846080307
  dataset_size: 21289144155.152
tags:
- japanese
- anime
- speech
- 日本語
- audio-text
- asr
- whisper
- voice
- large-v3
- not-for-all-audiences
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---

# Japanese Anime Speech Dataset V2

[**日本語はこちら**](https://huggingface.co/datasets/joujiboi/japanese-anime-speech-v2/blob/main/README_JA.md)

**japanese-anime-speech-v2** is an audio-text dataset designed for training automatic speech recognition models.
The dataset comprises **300,506 audio clips** and their corresponding transcriptions from various visual novels.
This dataset is <b>not</b> an expanded version of <a href="https://huggingface.co/datasets/joujiboi/japanese-anime-speech-v1">japanese-anime-speech-v1</a>.
For that reason, much of the audio from japanese-anime-speech-v1 is **not** included in this dataset.

The goal of this dataset is to increase the accuracy of automatic speech recognition models, 
such as OpenAI's [Whisper](https://huggingface.co/openai/whisper-large-v2), in accurately transcribing dialogue from anime and other similar Japanese media.
This genre is characterized by unique vocal and linguistic features and speech patterns that diverge from conventional Japanese speech.

<div class="course-tip course-tip-orange bg-gradient-to-br dark:bg-gradient-to-r before:border-orange-500 dark:before:border-orange-800 from-orange-50 dark:from-gray-900 to-white dark:to-gray-950 border border-orange-50 text-orange-700 dark:text-gray-400">
  <p>
    <b>Content Warning:</b> Please be advised that the audio in this dataset is sourced from visual novels and contains content that is not suitable for all audiences.
    <br>
    Unlike <a href="https://huggingface.co/datasets/joujiboi/japanese-anime-speech-v1">japanese-anime-speech-v1</a>,
    this dataset has <b>not</b> been filtered for NSFW content. If you use this dataset to train an automatic speech recognition model you *should* clean the dataset first.
  </p>
</div>

A list of all audio files and transcriptions can be found [**here**](https://huggingface.co/datasets/joujiboi/japanese-anime-speech-v2/blob/main/audio_transcription_list.txt).

# Dataset information

* **300,506** audio-text pairs
* **458.55 hours** of audio (OpenAI suggests a minimum of [5 hours](https://huggingface.co/blog/fine-tune-whisper) for productive [Whisper](https://huggingface.co/openai/whisper-large-v2) fine-tuning)
* **5.5s** average audio length
* Dataset source: **visual novels**
* Audio format: **mp3 (128kbps)**
* Latest version: **V1 - June 29th 2024** (this means the first version of the japanese-anime-speech-v2 series)

# Changes from V1

* Significantly increased dataset size from 16,143 to 300,506 audio-text pairs
* Changed audio from mp3 (192kbps) -> **mp3 (128kbps)** for better storage efficiency
* No filtering of NSFW content in this version - in the future, the dataset may be split into safe-for-work and nsfw on different splits

# Bias and Limitations

This dataset, while valuable for training anime-style Japanese speech recognition, has some inherent biases and limitations.
The audio is primarily sourced from visual novels, leading to a gender bias towards female voices and a domain-specific vocabulary revolving around topics such as love, relationships, and fantasy.
Additionally, the professionally produced nature of the audio results in clear and slow speech, which may not fully reflect real-world speaking patterns.
This version includes NSFW content, which is not suitable for all applications.
No formatting or cleaning of the transcriptions has been performed (e.g. removing duplicate characters), which may affect the quality of some text samples.

# Future Plans

* Create separate splits for safe-for-work and NSFW content
* Implement text formatting to improve transcription quality (e.g. removing excessive character repetition)
* Continue expanding the dataset to include more sources

# Use & Credit

This dataset is openly available for commercial or non-commercial use. Anyone is welcome to use this dataset as they deem appropriate.
However, the creator assumes no responsibility for the consequences of its use.
While not mandatory, crediting this dataset with a hyperlink in any derivative work would be greatly appreciated.

I hope that by sharing this dataset, we (the open-source community) improve automatic speech recognition for anime content.