Mandarin-English Code-Switching Dialogue Speech Recordings

CS-Dialogue is a 104-hour dataset of spontaneous Mandarin-English code-switching dialogues for speech recognition. It was created by BAAI to address limitations in existing datasets, such as small size and lack of natural conversations.

Use Cases

• Train automatic speech recognition models on 104 hours of spontaneous Mandarin-English code-switching dialogue audio.
• Analyze code-switching patterns and linguistic features within natural, full-length dialogue recordings.
• Develop language identification models for segments within bilingual speech dialogues.

Strengths

• Contains 104 hours of speech data, providing substantial audio for model training.
• Focuses on spontaneous and natural dialogue recordings, unlike scripted datasets.
• Publicly available and designed to address key gaps in code-switching speech research.

Limitations

• The dataset is specific to Mandarin-English code-switching, limiting generalizability to other language pairs.
• Sample data and detailed column structure are not provided in the input, making initial assessment difficult.
• Potential for demographic or accent bias exists within the spontaneous speaker pool.

Provenance

Source

BAAI

Collection Method

Collection of spontaneous speech dialogues.

Freshness

Last updated on 2025-07-22.