Most bilingual people do not speak one language at a time. Mid-sentence an English word simply lands better than the alternative, so that is the word you reach for. Speech technology still handles that badly, because almost no recorded data captures it. Only people who genuinely live between two languages can supply it.

You will join an online call with 1-2 other participants and record a natural conversation in your native language, mixing in English words and phrases the way you normally would. This is known as code-switching. Roughly one out of every three sentences should include English, reflecting real-world bilingual communication. Conversation topics can range from news events and personal interests to fictitious stories, keeping the dialogue authentic and engaging.

Each session involves a minimum of four recordings, ensuring ample data for model training. You will need a stable internet connection, a quiet environment, and a device with a working microphone. Sessions are conducted remotely, offering flexibility to contribute from anywhere.

This project suits native speakers who are also proficient in English and comfortable conversing naturally in both languages. Your participation directly supports speech technology that serves multilingual communities more accurately.

This project helps speech technology understand and process code-switching, a common pattern in multilingual communities. By providing authentic examples of mixed-language conversation, you enable speech recognition and natural language understanding models to serve the communities you are part of far better. The result is more inclusive, accurate, and context-aware language technology across global markets.

Work arrangement: Fully remote

Applicant eligibility: 8 eligible countries

One-off paid session per recorded conversation, paid on completion and quality acceptance. The session rate covers a full session of 3.5 hours. A shorter session is prorated against that, so you are paid for the time you take part.