Speech Foundation Models and Crowdsourcing for Efficient, High-Quality Data Collection