diff --git a/model/lyrics_parse.go b/model/lyrics_parse.go
index 3ccb3646a..175f964ae 100644
--- a/model/lyrics_parse.go
+++ b/model/lyrics_parse.go
@@ -19,49 +19,36 @@ func parseLyricsfileBytes(_ string, contents []byte) (LyricList, error) {
 	return parseLyricsfile(string(contents))
 }
 
-// parseTTMLIfDocument parses TTML only when the content actually looks like a
-// TTML document, so content-sniffing plain text or LRC does not run the XML
-// decoder (and never logs a spurious parse warning for non-TTML input).
-func parseTTMLIfDocument(lang string, contents []byte) (LyricList, error) {
-	if !isTTMLDocument(string(contents)) {
-		return nil, nil
-	}
-	return parseTTMLWithDefaultLang(lang, contents)
-}
-
-// registry is the single source of truth for the structured formats. Slice
-// order is the content-sniff probe order (TTML → SRT → YAML), and each row's
-// suffixes drive sidecar dispatch. byContent is used when sniffing — it gates a
-// format so, e.g., TTML only runs the XML decoder on something that looks like a
-// document; bySuffix parses unconditionally because the file extension already
-// declares the format. LRC/plain is not listed: it is the fallback floor for
-// both paths.
-var registry = []struct {
-	suffixes  []string
-	bySuffix  lyricParser
-	byContent lyricParser
+// lyricFormats lists the structured formats. Each parser is self-skipping: it
+// returns an empty list (not an error) when the content is not its format, so it
+// is safe to try in order. Slice order is the content-sniff probe order
+// (TTML → SRT → YAML); each row's suffixes drive sidecar dispatch. LRC/plain is
+// not listed — it is the fallback floor for both paths.
+var lyricFormats = []struct {
+	suffixes []string
+	parse    lyricParser
 }{
-	{[]string{".ttml"}, parseTTMLWithDefaultLang, parseTTMLIfDocument},
-	{[]string{".srt"}, parseSRTWithLanguage, parseSRTWithLanguage},
-	{[]string{".yaml", ".yml"}, parseLyricsfileBytes, parseLyricsfileBytes},
+	{[]string{".ttml"}, parseTTMLWithDefaultLang},
+	{[]string{".srt"}, parseSRTWithLanguage},
+	{[]string{".yaml", ".yml"}, parseLyricsfileBytes},
 }
 
 // ParseLyrics is the single entry point for parsing lyrics. A known suffix
 // (.ttml/.srt/.yaml/.yml/.lrc) routes to that format's parser; an empty or
-// "auto" suffix content-sniffs. In both modes a structured parser that does not
-// match falls back to the LRC/plain-text floor — never to another structured
-// format.
+// "auto" suffix content-sniffs by trying each format in order. In both modes a
+// structured parser that does not match falls back to the LRC/plain-text floor —
+// never to another structured format.
 func ParseLyrics(suffix, lang string, contents []byte) (LyricList, error) {
 	if suffix == "" || strings.EqualFold(suffix, "auto") {
-		candidates := make([]lyricParser, len(registry))
-		for i, r := range registry {
-			candidates[i] = r.byContent
+		candidates := make([]lyricParser, len(lyricFormats))
+		for i, f := range lyricFormats {
+			candidates[i] = f.parse
 		}
 		return parseFirstMatch(lang, stripBOM(contents), candidates...)
 	}
-	for _, r := range registry {
-		if slices.ContainsFunc(r.suffixes, func(s string) bool { return strings.EqualFold(s, suffix) }) {
-			return parseFirstMatch(lang, contents, r.bySuffix)
+	for _, f := range lyricFormats {
+		if slices.ContainsFunc(f.suffixes, func(s string) bool { return strings.EqualFold(s, suffix) }) {
+			return parseFirstMatch(lang, contents, f.parse)
 		}
 	}
 	return plainLRC(lang, contents) // .lrc and any unknown suffix
diff --git a/model/lyrics_ttml.go b/model/lyrics_ttml.go
index 4f49c8818..59686ce16 100644
--- a/model/lyrics_ttml.go
+++ b/model/lyrics_ttml.go
@@ -109,6 +109,14 @@ func parseTTML(contents []byte) (LyricList, error) {
 func parseTTMLWithDefaultLang(defaultLang string, contents []byte) (LyricList, error) {
 	contents = xmlEncodingRegex.ReplaceAll(contents, []byte(`<?xml$1encoding="UTF-8"$2?>`))
 
+	// Self-skip non-TTML content (returns nil, nil) so content-sniffing does not
+	// run the XML decoder on plain text or LRC. Checked after the encoding fixup
+	// above so a UTF-16-declared document is still recognized. Content that is a
+	// <tt> document but malformed still parses below and surfaces its error.
+	if !isTTMLDocument(string(contents)) {
+		return nil, nil
+	}
+
 	p := ttmlParser{
 		decoder: xml.NewDecoder(bytes.NewReader(contents)),
 		params: ttmlTimingParams{