Core demands
With the booming demand for content production, live broadcast scenes pose great challenges to the effective monitoring of live voice content due to the characteristics of strong real-time and large scale. YY has tens of thousands of real-time broadcasts every day. It is difficult to monitor the voice interaction information of the anchor and Lianmai users. When there is illegal chat content involving discussion of politics, pornography, malicious advertising, etc., in addition to user reports, suspicious user portrait detection, voice recognition technology can be used to transcribe the voice in the chat into text, Then conduct sensitive keyword detection, assist auditors to find suspicious information in time and follow up.
Solution
YY uploads the chat voice information in the live broadcast to the server in real time, and then calls Baidu voice recognition service to recognize the audio content as text in real time and return it. The auditor extracts keyword information from it, and tracks, verifies, and processes the suspicious live broadcast.