datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
turn-end-detection
Turn-end detection from real ASR prefixes, with audio
100,348 labelled end-of-turn decision points over
53,140 synthesized customer-service utterances, each one paired
with the 16 kHz audio it was cut from, plus the endpointing decisions
6 commercial endpointer configurations
made on the same audio.
The question each row poses is the one a voice agent has to answer continuously:
given everything heard so far, has the caller finished speaking? Ending the
turn too early talks over… See the full description on the dataset page: https://huggingface.co/datasets/mazesmazes/turn-end-detection.customer-turn-detection-rate-limit-safe-with-artifactsturn-detection-labeled-v3
🇷🇺 Russian Real-Estate Turn Detection (Probability Balanced)
This dataset is designed for training probability-based turn detection models for Russian conversational AI, specifically in the real-estate domain (renting, buying, inquiries).
It follows a contrastive approach: for every complete user query, there is a corresponding incomplete version. This forces the model to learn the subtle semantic and grammatical cues that signal the end of a turn versus a mid-sentence pause.… See the full description on the dataset page: https://huggingface.co/datasets/RAS1981/turn-detection-labeled-v3.turn-detection-labeled-v2customer-turn-detection-balanced-with-artifactscustomer-turn-detection-test-with-artifactscustomer-turn-detection-structured-with-artifacts