Case Study: Castbox achieves accurate, low-latency in-audio search and 50K concurrent users with Google Cloud Platform

CastBox Giving users a voice with Google Cloud

Castbox, founded in 2016 and headquartered in Beijing, is a global spoken-audio platform (50 million podcasts, 70 languages, 175 countries) serving about 2 million daily users. As the user base grew, Castbox needed a highly available, low-latency infrastructure to support up to 50,000 concurrent connections, deliver features like in-audio search, and warehouse and analyze rapidly growing data (about 30 GB/day) to personalize recommendations and iterate on the app.

By consolidating on Google Cloud Platform and Firebase—using Cloud Speech-to-Text, BigQuery, Firebase Realtime Database and A/B Testing—Castbox achieved >96% speech-to-text accuracy, ~50 ms search-query latency and ~250 ms audio-to-text conversion, processed roughly 20 hours of audio per day, and supported 50K concurrent connections with ~10 ms latency. The solution also enables large-scale analytics (queries up to ~1,200 GB), personalized recommendations, rapid A/B-driven product updates, and the ability to create original content based on user insights.

CastBox

Shibin Li

Co-founder