Majestrino Unified Detailed Captions: 4.6 Million Audio-Text Pairs

Majestrino Unified Detailed Captions is a filtered subset of the laion/majestrino-data collection, containing all samples with a unified_detailed_caption field. The dataset comprises 4,658,407 samples, packaged in approximately 932 tar files totaling around 1,017 GB. It was created by TTS-AGI and last updated on March 29, 2026.

Use Cases

• Training text-to-speech synthesis models based on the detailed captions paired with audio files.
• Fine-tuning automatic speech recognition systems based on the provided speech transcriptions.
• Developing multimodal audio classification models based on the labeled audio-caption pairs.
• Researching audio captioning and description generation based on the unified detailed captions.

Strengths

• Contains 4,658,407 samples, indicating a large-scale collection.
• Provides paired audio (.flac) and metadata (.json) files, ensuring data alignment.
• Includes multiple text fields such as 'caption', 'transcription', and 'duration', offering varied annotations.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

TTS-AGI, filtered from laion/majestrino-data.

Collection Method

Filtered subset containing all samples with the 'unified_detailed_caption' field.

Time Range

null

Freshness

Last updated 2026-03 29 08:25:34; freshness should be verified.

Geography

null