100 Japanese Female Voice Clones from ITA-Corpus Emotion

100 female voice audio clips generated using Qwen3-TTS, based on the public-domain ITA-Corpus Emotion text dataset. The audio is provided in 24kHz mono WAV format, with each voice having a descriptive label.

Use Cases

• Train a voice cloning model using the 100 distinct WAV files as target audio samples.
• Analyze acoustic features like pitch and timbre across the 100 labeled voice profiles (e.g., 'lower-pitched female voice with a strong core').
• Benchmark text-to-speech synthesis systems using the phonetically balanced ITA-Corpus Emotion text source.
• Develop emotion or style transfer models by leveraging the descriptive labels assigned to each voice.

Strengths

• Contains 100 distinct female voice samples, providing a fixed scale for experiments.
• Audio is consistently formatted as 24kHz mono WAV files.
• Based on the public-domain ITA-Corpus Emotion dataset, ensuring legal clarity for the text source.

Limitations

• Limited to 100 samples, which is a small dataset for training large-scale generative models from scratch.
• Only contains Japanese female voices, lacking demographic and linguistic diversity.
• The quality and naturalness of the Qwen3-TTS generated audio are not independently verified.

Provenance

Source

Generated from the ITA-Corpus Emotion text dataset using Qwen3-TTS.

Collection Method

Voice synthesis via the Qwen3-TTS model.

Time Range

null

Freshness

Last updated on 2026-02-01.

Geography

Japan (Japanese language).