Nyan is a Japanese text-to-speech dataset generated by Irodori-TTS-600M-v3-VoiceDesign and formatted for fine-tuning the LFM2.5-Audio-1.5B model. It contains 9,322 training samples and 491 validation samples. The dataset was authored by RikkaBotan and last updated on June 6, 2026.
Use Cases
- Fine-tuning audio generation models based on the dataset's structured prompts and text samples.
- Training Japanese speech synthesis systems based on the dataset's language focus.
- Evaluating text-to-speech model performance using the provided validation split.
- Studying voice style transfer based on the system_prompt column.
- Generating synthetic speech data for downstream applications.
Strengths
- Dataset includes 9,322 training samples, providing a substantial base for model training.
- Contains a dedicated validation split of 491 samples for model evaluation.
- Columns suggest structured metadata including voice style instructions and text inputs.
Limitations
- Description metadata is limited; actual data quality requires manual inspection after download.
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
Provenance
- Source
- RikkaBotan
- Collection Method
- Generated by Irodori-TTS-600M-v3-VoiceDesign.
- Freshness
- Last updated 2026-06-06 04:44:34; freshness should be verified.