Ocular AI

Ocular AI sells human collected and human annotated training data for speech and multimodal AI models. Its catalog includes speech to text audio data, studio grade text to speech recordings, full duplex conversational audio and full duplex audiovisual data, in which two speakers are recorded on separate streams so overlap, interruption and backchannels are preserved. Each recording ships with human verified transcripts, speaker metadata and annotations for disfluencies and paralinguistics. A transcription service applies the same annotation pipeline to a customer's own audio. The conversational datasets span American English, French, Arabic, Spanish, Indonesian, Russian, Mandarin, Vietnamese, Japanese, Thai, German, Hindi, Korean and Polish, and sample sets can be requested for each. The audiovisual product pairs synced 4K video with audio for each participant, for video conversational models, interactive avatars and interaction models. Ocular AI also publishes the Converse STT Benchmark, a leaderboard that measures how accurately 15 frontier speech models transcribe real two person American English conversations. The buyers are AI labs and companies training or evaluating speech, voice and interaction models. The site lists a marketplace and a Get Data route, and it also runs an expert network that supplies annotation work. The vendor publishes no prices for datasets or services, and samples are requested through the site.