Streaming audio while encoding: how to avoid latency with ffmpeg

This article addresses the latency issue when streaming long audio files transcoded to Opus using ffmpeg. In a standard setup, the server waits for the entire file to be written, which causes a 19-second delay for a 46-minute recording. The author explores methods to stream data to the client in real-time while the encoding process is still ongoing. The article covers technical challenges, including why standard pipes are not always effective, how to avoid race conditions during data reads, the nuances of file renaming on Windows, and the importance of container fragmentation (e.g., M4A) for instant playback. The proposed solution optimizes server load by using a single ffmpeg pass for all requests while ensuring effective content caching.