UniST: Speech Token Training Data for Codec and Language Models

UniST codec-token training data was exported from local metadata and codec results. The dataset was created by author 'cmots' and was last updated on June 3, 2026. It is used to train the UniSS model.

Use Cases

• Train speech-to-text models based on source and target GLM token lists.
• Develop codec models based on bicodec semantic and global token lists.
• Build multilingual translation systems based on transcription and translation text fields.
• Benchmark tokenization methods for speech data using the provided token lists.

Strengths

• Includes multiple token representations (GLM, bicodec) for each sample.
• Contains parallel text data in the form of transcriptions and translations.
• Explicitly designed for training the UniSS model, indicating a specific application.

Limitations

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

Provenance

Source

Exported from local metadata and codec results.

Freshness

Last updated 2026-06-03 07:31:02; freshness should be verified.