01
What is Open Yap 1K?
Open Yap 1K is a dual-channel dataset of natural English two-speaker conversations released by The Agentic Data Company. The full release contains 1,000 hours across 1,602 conversations and 239 speakers.
Each speaker is recorded on a separate channel aligned to the same timeline. That preserves interruptions, overlap, backchannels, laughter and short turn-taking gaps instead of collapsing both sides into one mixed track.
02
Open Yap 1K specs
- 1,000 hours in the full corpus
- 1,602 conversations
- 239 unique speakers
- English two-speaker natural conversation
- channel-separated audio on a shared timeline
- word-level machine-generated transcripts
The publisher documents 48 kHz audio for the release. The full corpus is described as 16-bit PCM, while the public Hugging Face sample uses 48 kHz, 16-bit FLAC. The transcripts are generated with Deepgram Nova-3 and are not human-verified.
03
Why the dual-channel format matters
Full-duplex voice systems need to handle more than clean alternating turns. They must decide when to speak, when to wait, how to react to backchannels and what to do when a user interrupts.
Separate synchronized tracks make those behaviors directly observable. The publisher reports a median turn-taking gap of 580 milliseconds, median overlap equal to 8.3% of voiced time and a median 11.9 turns per minute; these remain vendor-reported corpus statistics.
04
Sample versus full-corpus access
Hugging Face hosts an 8.9-hour sample from 16 conversations and eight speakers. The sample is hand-picked rather than a random statistical draw from the full corpus and is listed under CC BY 4.0.
The complete 1,000-hour corpus is a separate request-based release. The Agentic Data Company says anyone may request access and that approved recipients receive either a direct download or delivery into their own S3 bucket under the Open Yap 1K Data Use Agreement.
05
Commercial use and licensing
The current full-corpus agreement permits commercial and research use, model training, fine-tuning, evaluation and deployment of trained models. The agreement is separate from the sample's CC BY 4.0 license.
06
Use cases and important limits
The release is relevant to full-duplex speech-to-speech systems, expressive text-to-speech, turn-taking research, endpointing and long-form audio understanding. The data format can help teams study overlapping speech without first reconstructing who spoke when from a mixed signal.
Open Yap 1K is English-only, its transcripts are machine-generated, the public sample is intentionally hand-picked and full-corpus access is request-gated. The dataset does not provide one universal model-training recipe, hardware requirement or benchmark guarantee.
07
Bottom line
Open Yap 1K is notable because its 1,000 hours preserve the conversational timing that full-duplex voice systems must learn: overlap, interruptions, backchannels and rapid turn-taking on separate aligned channels.
For the exact download paths, sample/full license split and current commercial-use restrictions, use the dedicated access guide linked below.
Sources
Primary and supporting sources
Facts were rechecked against the linked sources immediately before publication. Pricing, product availability and rollout status can change.