Large-Scale English Speech Corpus for Text-to-Speech Training

615,000 hours of English speech audio from 239.7 million segments, aggregated from 11 source datasets. The corpus was constructed by KRAFTON from 8 public speech corpora and web-sourced recordings to train the RAON-OpenTTS model. The dataset page was last updated in April 2026.

Use Cases

• Training TTS models using the aggregated 615,000 hours of speech audio and associated text transcripts.
• Benchmarking speech synthesis quality against the RAON-OpenTTS model trained on this corpus.
• Analyzing acoustic features and prosody across the 239.7 million speech segments from diverse sources.
• Pre-training or fine-tuning voice cloning systems using the large-scale, multi-source audio data.

Strengths

• Extremely large scale with 615,000 hours of speech audio.
• Aggregates data from 11 distinct source datasets for diversity.
• Contains 239.7 million individual speech segments.

Limitations

• Specific column names, data structure, and file formats are not detailed in the provided description.
• The geographic and demographic coverage of the aggregated sources is unspecified, which may introduce bias.
• The technical report detailing construction methodology is listed as 'Coming soon'.

Provenance

Source

Aggregated from 8 publicly available speech corpora and a set of web-sourced recordings.

Collection Method

Constructed by KRAFTON, combining 11 source datasets into a unified pool.

Freshness

Dataset page was updated on 2026-04-02.