Japanese Eroge Voice V2: 1-10 Million Audio-Transcription Pairs

1 to 10 million audio-transcription pairs extracted from Japanese adult games by NandemoGHS in January 2026. The dataset consists of entirely new audio clips and transcriptions with no overlap from the original version.

Use Cases

• Training Text-to-Speech (TTS) models using the Japanese transcriptions and audio clips
• Developing Automatic Speech Recognition (ASR) systems for Japanese anime-style dialogue
• Fine-tuning voice conversion models using the specific vocal characteristics found in eroge audio

Strengths

• Contains between 1,000,000 and 10,000,000 records
• Released under the MIT license for open research
• Zero overlap with the previous version (V1), providing unique data points

Limitations

• Domain-specific bias toward adult game dialogue and character archetypes
• Potential for explicit or NSFW linguistic content in transcriptions
• Lack of metadata regarding speaker demographics or recording environments

Provenance

Source

Japanese adult games (eroge)

Collection Method

Extracted from game files

Freshness

Last updated January 2026.

Geography

Japan