Emo-TTS: High-Arousal Emotional Speech Evaluation Datasets

Emo-TTS Evaluation Datasets are used for evaluating a training-free inference framework for high-arousal emotional speech synthesis. The collection includes the HIED dataset with 400 samples across four emotions and the ESD dataset with 20 speakers across two languages. The dataset was created by author 'erminga' and was last updated on 2026-04-29.

Use Cases

• Benchmarking emotional speech synthesis models based on the described high-arousal emotional datasets.
• Evaluating speaker and language generalization in TTS based on the multi-speaker, bilingual ESD dataset.
• Training or fine-tuning models for specific emotional categories like anger, happiness, sadness, and surprise based on the labeled emotions.

Strengths

• Includes a dedicated high-intensity emotional dataset (HIED) with 400 samples.
• Contains the established Emotional Speech Dataset (ESD) with 20 speakers and two languages (EN/ZH).
• Dataset is specifically designed for evaluating a novel, training-free TTS inference framework.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the overall collection is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality and file formats require manual inspection after download.

Provenance

Source

erminga on Hugging Face

Collection Method

Likely compiled for evaluating the Emo-TTS framework; HIED is described as 'ours'.

Freshness

Last updated 2026-04-29 05:43:17; freshness should be verified.