diff --git a/README.md b/README.md
index 0af4e3dd..8c1f3dfe 100644
--- a/README.md
+++ b/README.md
@@ -42,7 +42,7 @@ _LMS_ provides several ways to help you find the music you like:
 
 __Note__: the recommendation engine supports two modes:
 * **Tag-based**: uses metadata tags `genre` and `grouping`.
-* **Audio similarity**: uses MusicNN embeddings. Extracting embeddings takes roughly 10× longer than regular file metadata parsing. To speed up extraction, consider tweaking `scanner-thread-count` in `lms.conf` (defaults to half the number of logical CPUs).
+* **Audio similarity**: uses MusicNN embeddings. Extraction throughput depends heavily on both the number of scanner threads and the speed of your storage. Typical reported speeds range from ~1k tracks/hour on a Raspberry Pi 4 (3 threads) to ~25k tracks/hour on an Intel Core i5-13500 (10 threads). To speed up extraction, consider tweaking `scanner-thread-count` in `lms.conf` (defaults to half the number of logical CPUs).
 
 ## About tags
 _LMS_ primarily relies on tags to organize your music collection but also supports browsing by directory using the [Subsonic/OpenSubsonic API](SUBSONIC.md).