Babel Audio Pays Strangers $17 an Hour to Record Conversations for AI Training Data

Babel Audio, a startup building AI training datasets from human conversations, pairs anonymous strangers to record their discussions and bundles the recordings as AI training material, paying participants $17 per hour, according to a report by Issie Lapowsky at Bloomberg. The company operates in the data labeling and synthetic data market, where demand for high-quality human-generated audio has grown as AI labs race to train and fine-tune speech and conversational models. Unlike scripted recordings or synthetic audio generation, Babel Audio's approach captures naturalistic conversation patterns, including hesitations, topic shifts, and informal language, which are difficult to replicate artificially. The $17 per hour rate positions the work above minimum wage in most U.S. markets, reflecting competition for participants willing to have their voices recorded and used commercially. Participants are matched anonymously, and conversations are recorded through the company's platform. The model touches on a broader set of questions about how AI companies source training data and whether participants in data generation pipelines are adequately informed and compensated. Ongoing litigation against major AI companies including OpenAI, Meta, and Stability AI has centered on whether the use of web-scraped content for training constitutes infringement, creating commercial incentive for companies to develop consented, compensated data pipelines. Babel Audio joins a growing market of paid data collection platforms including Scale AI, Appen, and Remotasks, which have built large workforces of human annotators, evaluators, and data generators. The distinction is that Babel focuses specifically on conversational audio. The company did not disclose its funding, investor base, or the volume of data it has collected to date.