318,729 speech samples annotated with 57 voice taxonomy dimensions on an ordinal scale from 0 to 6. Labels were generated by an ensemble of four Whisper models voting. The dataset was created by TTS-AGI and is designed as pre-training data for voice attribute classifiers.
Use Cases
- Pre-training voice attribute classifiers based on the 57 annotated dimensions.
- Benchmarking speech analysis models using the ensemble-generated ordinal labels.
- Developing multi-dimensional voice profiling systems based on the taxonomy.
- Studying the performance of label generation via model ensembles.
Strengths
- 318,729 speech samples provide a substantial volume of training data.
- 57 distinct voice taxonomy dimensions offer a multi-faceted annotation scheme.
- Labels generated by a four-model Whisper ensemble likely improve annotation consistency.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
- Last updated 2026-04-08 07:01:38; freshness should be verified.
Provenance
- Source
- TTS-AGI
- Collection Method
- Annotations generated by an ensemble of four Whisper models voting.