今度こそ解った気がする! chunk長とinputするあオーディオファイル長を揃えてるのがむしろ災いの元だったっぽ
“Audio file must match chunk size”❌ No — you can load long audio, chop into chunks
“Longer chunk = richer result”⚠️ Only if your model is large enough to handle it
“Short chunks repeat less”✅ They do — because they’re easier to diversify