AI Voices Deduplicated: 2,004 High-Quality Speaker-Deduplicated Audio Samples

2,004 high-quality AI voice samples derived from a larger collection of approximately 32,000 samples. The dataset was created by LAION through a process of quality filtering and speaker deduplication using speaker embeddings and clustering. It was last updated on March 17, 2026.

Use Cases

• Training text-to-speech models based on high-quality, filtered audio samples.
• Developing speaker verification systems based on deduplicated speaker embeddings.
• Benchmarking audio quality and speaker diversity in synthetic voice datasets.
• Fine-tuning voice cloning models on a curated set of distinct AI voices.

Strengths

• Contains 2,004 samples that passed quality filters for background quality and content enjoyment.
• Underwent speaker deduplication to reduce redundancy, likely improving speaker diversity.
• Derived from a substantial source pool of approximately 32,000 initial samples.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the final dataset is known, but the original feature set and file formats are unknown.
• Data may reflect source bias inherent to the original LAION collection.

Provenance

Source

laion/reference_ai_voices_with_timbre_annotations

Collection Method

Quality filtered (score_background_quality >= 3.5 AND score_content_enjoyment >= 5.0) and speaker deduplicated using embeddings and agglomerative clustering.

Freshness

Last updated 2026-03-17 15:03:32; freshness should be verified.