diff --git a/crates/media/src/audio.rs b/crates/media/src/audio.rs
index e3769ee2f..09f9efa37 100644
--- a/crates/media/src/audio.rs
+++ b/crates/media/src/audio.rs
@@ -81,7 +81,9 @@ pub fn decode(bytes: &Bytes) -> Result<Pcm> {
             Some(r) => r,
             None => contained(|| decode_symphonia(bytes, "mkv")),
         },
-        _ if !crate::mp4guard::probe_fits(bytes) => Err(MediaError::Corrupt("the MP4's sample tables do not fit the file".into())),
+        Container::Mp4 | Container::Unknown | Container::Mp3 if !crate::mp4guard::probe_fits(bytes) => {
+            Err(MediaError::Corrupt("the MP4's sample tables do not fit the file".into()))
+        }
         c => contained(|| decode_symphonia(bytes, c.name())),
     }
 }
diff --git a/crates/media/src/mp4guard.rs b/crates/media/src/mp4guard.rs
index 86ffc3528..56047979f 100644
--- a/crates/media/src/mp4guard.rs
+++ b/crates/media/src/mp4guard.rs
@@ -19,9 +19,18 @@ const PROBE_SEARCH_LIMIT: usize = 1024 * 1024;
 /// `ftyp` within its search window as the start of an MP4 (the size field 4 bytes before it), so
 /// the table check must start where that parser will, not at offset 0.
 pub(crate) fn probe_fits(b: &[u8]) -> bool {
-    let head = &b[..b.len().min(PROBE_SEARCH_LIMIT + 16)];
+    if crate::Container::sniff(b) == crate::Container::Mp4 && !tables_fit(b) {
+        return false;
+    }
+    // The probe reads leading ID3v2 tags as metadata and searches again after the last one.
+    let mut rest = b;
+    while rest.starts_with(b"ID3") {
+        let size = rest.get(6..10).map(|s| s.iter().fold(0usize, |a, &x| (a << 7) | (x & 0x7f) as usize)).unwrap_or(0);
+        rest = rest.get(10 + size..).unwrap_or(&[]);
+    }
+    let head = &rest[..rest.len().min(PROBE_SEARCH_LIMIT + 16)];
     match head.windows(4).position(|w| w == b"ftyp") {
-        Some(p) => tables_fit(&b[p.saturating_sub(4)..]),
+        Some(p) => tables_fit(&rest[p.saturating_sub(4)..]),
         None => true,
     }
 }
diff --git a/crates/media/src/probe.rs b/crates/media/src/probe.rs
index 4de516b1f..4e8e845c5 100644
--- a/crates/media/src/probe.rs
+++ b/crates/media/src/probe.rs
@@ -71,8 +71,8 @@ pub fn probe_headers(bytes: &Bytes) -> Result<MediaInfo> {
     };
     if info.audio.is_none() && info.video.is_none() || info.audio.as_ref().is_some_and(|a| a.sample_rate == 0) {
         // Plain audio files (and anything our demuxers didn't describe): symphonia's probe.
-        // Known non-MP4 containers (a WAV's audio bytes can spell an `ftyp`) skip the MP4 table check.
-        if matches!(c, Container::Mp4 | Container::Unknown) && !crate::mp4guard::probe_fits(bytes) {
+        // Symphonia opens an MP4 wherever its probe finds an `ftyp` (after leading ID3v2 tags too); other known containers (a WAV's audio bytes can spell an `ftyp`) skip the check.
+        if matches!(c, Container::Mp4 | Container::Unknown | Container::Mp3) && !crate::mp4guard::probe_fits(bytes) {
             return Err(MediaError::Corrupt("the MP4's sample tables do not fit the file".into()));
         }
         let (codec, rate, channels, frames) = std::panic::catch_unwind(|| crate::audio::probe_symphonia(bytes, c.name()))