ThBel/seamless-interaction-transcripts

Seamless Interaction Transcripts

Dataset Summary

Seamless Interaction Transcripts is a large-scale dialogue dataset derived from facebook/seamless-interaction dataset. It contains verbatim transcriptions of >3k dialogues in English covering a range of contexts from general chit-chat to customer service.

The dataset is designed to support research and development of speech and dialogue systems that require modeling of conversational turn-taking, such as real-time conversational agents.

Source Data

• —Base datasets:
• —facebook/seamless-interaction
• —Language(s): English
• —Modality: Text
• —Interaction type: Human conversational speech
• —Dialogues: 3335
• —Avg. Turns / Dialogue: 37.47
• —Speakers: 2000+

Dataset splits (e.g., train/validation/test) are not predefined and may be created by downstream users as needed. Deduplication was applied to the underlying text sources to ensure dataset splits can be made without contamination.

Annotation Details

• —Transcripts

• —Generated automatically using Whisper Large V3 Turbo.
• —A subset of samples (\~200) was manually reviewed. The transcripts are estimated to have approximately a word error rate (WER) of \~2.8%.

Dataset Structure

A typical data entry includes:

• —interaction_id: Identifier for the interaction
• —messages: Transcriptions in the form of an OpenAI-compatible chat message buffer
• —relation: High level relationship identifier, e.g. "strangers", "familiar"
• —relation_detailed: Detailed relationship identifier, e.g. "coworkers", "spouse"

Usage

In order to load the dataset from the hub, you can use the datasets library:

ds = datasets.load_dataset(
    "ThBel/seamless-interaction-transcripts",
    split='train',
    streaming=True # (optional)
)

for row in ds:
    # Do something with the data
    print(row['messages'])

Alternatively you may clone the ThBel/seamless-interaction-transcripts repository, and load the underlying parquet files using pandas.read_parquet.

Disclaimer and Licensing

Note that Seamless Interaction Transcripts is a derived dataset. I am not the original creator of the source dataset (facebook/seamless-interaction) and hold no rights over its content. This dataset is provided as-is for research purposes, and all credit goes to the original authors.

Annotations are released under the CC-BY-NC 4.0 license and are intended to be compatible with the licensing terms of the source dataset.

Citation

If you use the Seamless Interaction Transcripts dataset in academic, non-commercial work, please reference the original dataset.