Reazon Speech V2 Denoised: Japanese Speech Audio with Background Noise Removed

3,674 denoised audio files from the Reazon Speech v2 dataset, processed using UVR to remove background music and noise. The dataset was cleaned by author Stardust-minus using eight A800 GPUs over approximately 10 days and was mirrored to Hugging Face by litagin in April 2024.

Use Cases

• Train automatic speech recognition (ASR) models on denoised Japanese speech.
• Benchmark audio enhancement algorithms against a pre-processed dataset.
• Fine-tune voice synthesis models using high-quality, noise-free speech samples.

Strengths

• 3,674 processed audio files represent a substantial subset of the original Reazon Speech v2 corpus.
• Processing involved eight high-performance A800 GPUs, suggesting significant computational effort was invested in cleaning.

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.

Provenance

Source

Reazon Speech v2 dataset, processed by Stardust-minus.

Collection Method

Audio files denoised and had background music removed using UVR.

Freshness

Last updated 2024-04-16 23:17:49; freshness should be verified.