Talker-T2AV: Clean Training Data for Joint Talking Audio-Video Generation

Talker-T2AV-Data is a clean training data package for the Talker-T2AV model, which performs joint talking audio-video generation using autoregressive diffusion modeling. The dataset is hosted by HKUSTAudio and was last updated on 2026-05-24. It contains metadata and grouped archive shards for audio, motion, and video modalities.

Use Cases

• Train autoregressive diffusion models for joint audio-video generation based on the described multimodal data.
• Benchmark multimodal synthesis techniques using the clean, grouped training shards mentioned in the description.
• Develop talking head or audio-driven animation systems based on the motion and video data referenced.

Strengths

• Data is explicitly described as 'clean training data' for the model.
• Dataset is structured with grouped archive shards for different modalities (audio, motion, video).

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 and dataset scale are unknown, which may limit suitability assessment.

Provenance

Source

HKUSTAudio

Collection Method

Likely gathered and cleaned as training data for the Talker-T2AV research model.

Freshness

Last updated 2026-05-24 05:42:45.