EN Emilia Yodas 616H: English Speech Audio with Event Classification

616 hours of English speech audio derived from the Emilia-Dataset. The audio events are classified using the Scribe v1 tool, which employs ElevenLabs' speech-to-text technology. The dataset is a version 1 release by MrDragonFox, last updated on April 14, 2025.

Use Cases

• Train automatic speech recognition (ASR) models based on the English audio content.
• Develop audio event classification systems based on the Scribe v1 tool classifications.
• Filter audio data for quality using the mentioned Facebook audio aesthetics prefilter.
• Benchmark speech-to-text systems on a large-scale English audio corpus.

Strengths

• Contains 616 hours of English speech audio, providing substantial material for model training.
• Audio events are classified using a specific toolchain (Scribe v1 with ElevenLabs STT/ASR).
• The dataset is based on the Emilia-Dataset, which is licensed under CC BY 4.0.

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.
• The dataset is described as 'very much at a v1', suggesting it may be an early, potentially unrefined release.

Provenance

Source

Derived from the English part of the Emilia-Dataset on Hugging Face.

Collection Method

Audio event classification performed via Scribe v1 (ElevenLabs STT/ASR).

Freshness

Last updated 2025-04-14 21:23:31; freshness should be verified.