Skip to content
Sign in
Artwork for My Weird Prompts
My Weird Prompts · September 26 · 25 min

Audio Post-Processing for AI Podcast Pipelines

The front end of an AI podcast pipeline gets all the attention — agents, prompts, retrieval. But the stage nobody films a documentary about is audio post-processing, and it's where the actual craft lives. This episode walks through four specific edits: silence truncation, loudness normalization, EQ, and speed normalization without pitch distortion. Along the way, we dig into why SoX and FFmpeg split the work, why a single wrong integer in FFmpeg's silenceremove filter can silently delete 90% of your episode, and why the order of operations between silence removal and loudnorm isn't a style preference — it's a correctness issue. Plus: when to preview with silencedetect, how EBU R128's double-pass mode works, and whether any of these jobs actually benefit from GPU inference. Episode #411065 — open it directly at myweirdprompts.com/411065

0:00 · Intro-25:58
Play in
Transcript file

transcript

No transcript — this publisher did not publish one.

show notes

The front end of an AI podcast pipeline gets all the attention — agents, prompts, retrieval. But the stage nobody films a documentary about is audio post-processing, and it's where the actual craft lives. This episode walks through four specific edits: silence truncation, loudness normalization, EQ, and speed normalization without pitch distortion. Along the way, we dig into why SoX and FFmpeg split the work, why a single wrong integer in FFmpeg's silenceremove filter can silently delete 90% of your episode, and why the order of operations between silence removal and loudnorm isn't a style preference — it's a correctness issue. Plus: when to preview with silencedetect, how EBU R128's double-pass mode works, and whether any of these jobs actually benefit from GPU inference.

Episode #411065 — open it directly at myweirdprompts.com/411065

chapters

5 chapters

more episodes

All episodes