From 42658b3cbfcf3059ded94da6438ba9dfe6f70cc8 Mon Sep 17 00:00:00 2001
From: Honey <me_honeyy@proton.me>
Date: Sat, 2 May 2026 12:46:20 +0200
Subject: [PATCH] fix(backend): improve yt-dlp match filters for accuracy

- artistFilterWord: use longest \w+ run instead of stripping special chars,
  fixes artists like N!GHT where stripping gave NGHT (not a substring)
- buildTitleKeyword: new fn extracts most distinctive title word, added as
  title *= filter to all cascade steps to block unrelated fallback downloads
- Mix version specificity: Radio/Extended/Club Mix titles also filter on the
  qualifier word so plain versions are rejected
- Switch uploader *= to artist *= so label-uploaded tracks (Monstercat etc.)
  match by music metadata artist tag, not channel name

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---
 backend/novasonic-server.js | 40 +++++++++++++++++++++++++++++--------
 1 file changed, 32 insertions(+), 8 deletions(-)

diff --git a/backend/novasonic-server.js b/backend/novasonic-server.js
index 9486b4f..e3c73d8 100644
--- a/backend/novasonic-server.js
+++ b/backend/novasonic-server.js
@@ -56,13 +56,35 @@ function sanitize(s) {
   return (s || 'Unknown').replace(/[\\/:*?"<>|]/g, '_').trim().slice(0, 100);
 }
 
-// Extract a plain ASCII-ish word from an artist name for yt-dlp --match-filter uploader check.
+// Extract a plain word from an artist name for yt-dlp --match-filter uploader check.
+// Uses longest contiguous \w+ run so special chars like "N!GHT" → "N" + "GHT" picks "GHT",
+// but actually picks longest: "N!GHT" → runs ["N","GHT"] → "GHT" (3 chars).
 // "☆NoAki☆" → "NoAki", "Chiru-san" → "Chiru", "A, B" → "A"
 function artistFilterWord(name) {
   const first = (name || '').split(/[,/•]/)[0].trim();
-  const clean = first.replace(/[^\w\s-]/g, '').trim();   // strip ☆ etc.
-  const word  = clean.split(/[\s-]+/)[0];                // first word/segment
-  return word.length >= 3 ? word : '';
+  const runs = first.match(/\w+/g) || [];
+  const best = runs.filter(r => r.length >= 3).sort((a, b) => b.length - a.length)[0];
+  return best || '';
+}
+
+// Extract the most distinctive word from a track title to use as a title relevance filter.
+// Prevents fallback steps from downloading completely unrelated tracks.
+// Also preserves version qualifiers ("radio mix", "extended mix", "club mix", "vip mix") so
+// "Eclipse - Radio Mix" filters for "eclipse" AND "radio" to avoid downloading the plain version.
+function buildTitleKeyword(title) {
+  const VARIANT_WORDS = new Set(['sped','speedup','slowed','reverb','nightcore','karaoke','parody','remix','cover','acoustic','instrumental']);
+  const STOPWORDS = new Set(['the','a','an','and','or','of','in','on','at','to','for','feat','ft','with']);
+  const MIX_QUALIFIERS = ['radio', 'extended', 'club', 'vip', 'original', 'edit'];
+  const lower = (title || '').toLowerCase();
+  const words = lower
+    .replace(/[^\w\s]/g, ' ')
+    .split(/\s+/)
+    .filter(w => w.length >= 3 && !VARIANT_WORDS.has(w) && !STOPWORDS.has(w));
+  const main = words.sort((a, b) => b.length - a.length)[0] || '';
+  // If the title has a mix qualifier AND "mix", include the qualifier word too
+  const qualifier = lower.includes('mix') ? MIX_QUALIFIERS.find(q => lower.includes(q)) : null;
+  if (qualifier && qualifier !== main) return `${main} & title *= ${qualifier}`;
+  return main;
 }
 
 // Build yt-dlp --match-filter conditions to reject variant versions (speed up, slowed, remix…)
@@ -261,15 +283,17 @@ async function downloadTrack({ jobId, artist, albumArtist, title, album, year, t
 
   const uploaderWord   = artistFilterWord(mainArtist);
   const variantFilter  = buildVariantFilter(title);
-  const uploaderPart   = uploaderWord  ? ` & uploader *= ${uploaderWord}` : '';
+  const titleKeyword   = buildTitleKeyword(title);
+  const uploaderPart   = uploaderWord  ? ` & artist *= ${uploaderWord}` : '';
   const variantPart    = variantFilter ? ` & ${variantFilter}` : '';
+  const titleKwPart    = titleKeyword  ? ` & title *= ${titleKeyword}` : '';
 
-  // Build a --match-filter arg from active parts (duration, artist, variant).
-  // !is_live is always included.
+  // Build a --match-filter arg from active parts (duration, artist, variant, title keyword).
+  // !is_live is always included. titleKw always included to avoid totally unrelated downloads.
   const mf = (dur, artist, variant) => {
     const durPart = dur ? `duration > ${dur.min} & duration < ${dur.max} & ` : '';
     return ['--match-filter',
-      `${durPart}!is_live${artist ? uploaderPart : ''}${variant ? variantPart : ''}`];
+      `${durPart}!is_live${titleKwPart}${artist ? uploaderPart : ''}${variant ? variantPart : ''}`];
   };
 
   if (durationMs) {