FTSpeech-PNC-DA: Danish Punctuation and Capitalization Text for Speech Utterances

RyeAI/ftspeech-pnc-da is a text-only companion dataset for Danish speech processing. It provides restored punctuation and capitalization for utterances from the original FTSpeech dataset, with each row containing an utterance_id for joining back to the source. The dataset was created by RyeAI and last updated on 2026-06-19.

Use Cases

• Fine-tuning punctuation restoration models based on the described text pairs.
• Improving capitalization prediction for Danish ASR outputs based on the restored text.
• Training or evaluating text normalization pipelines for Danish speech-to-text applications.
• Creating synthetic training data for Danish NLP tasks that require proper formatting.

Strengths

• Dataset is explicitly derived from the established alexandrainst/ftspeech dataset, suggesting a known origin.
• Each row includes an utterance_id, which likely enables precise alignment with the original speech data.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for large-scale training.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Derived from the Hugging Face dataset alexandrainst/ftspeech.

Collection Method

Likely generated through automated or manual text restoration processes applied to the original speech transcriptions.

Freshness

Last updated 2026-06-19 09:10:00; freshness should be verified.