AUDIODIALOGUE
Natural English Dialogue Sets
Natural multi-speaker conversation with the turn structure, timing and speaker context needed for ASR, diarisation and conversational systems.
View datasetAUDIOACCENT COVERAGE
Accent and Regional Speech Sets
A broad language label can hide the variation that actually affects model behaviour. Accent and Regional Speech Sets make selected speaker groups explicit in the dataset, allowing regional or accent variation to remain visible rather than disappearing into a single speech corpus.
This is useful when a team needs to understand where a speech system performs differently and why. Speaker-group metadata can be connected to the same transcript and recording structure used across the collection, making comparisons easier without treating accent as the only property of the speaker.
Technical structure
- Speaker profiles with project-defined accent, region or locale fields
- Transcript and prompt identifiers linked to each utterance
- Device and recording-condition metadata where relevant
- Group structure designed around the comparisons the project needs to support
Typical applications
ASR adaptation, accent coverage, speech robustness, voice-agent development and group-level error analysis.
Delivery note
Target groups and metadata definitions are agreed for the collection rather than inferred from a generic taxonomy.
Delivery
The exact package depends on the collection scope. Where relevant, delivery can include task-specific records, manifests, provenance fields, stable identifiers, SHA-256 hashes, MLCommons Croissant 1.0 metadata and loading instructions.