Blend real sonic distance into classic getSimilarSongs ranking

similarSongsFor still filters by genre (unchanged, and getSonicSimilarTracks
remains the pure-sonic endpoint), but when the seed song has been
analyzed, genre-matched candidates that also have stored audio_features
are now ordered by actual feature-space distance from the seed instead
of SQL's random order; candidates without analysis data are left at
the end in their existing relative order rather than dropped or given
a fabricated distance.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014r98AtcBbevpscWLsfwVvv
This commit is contained in:
2026-08-23 08:31:54 -05:00
co-authored by Claude Sonnet 5
parent b6acd63c2d
commit 2a69f6dc33
2 changed files with 155 additions and 4 deletions
+64 -4
View File
@@ -4,10 +4,12 @@ import (
"database/sql"
"os"
"path/filepath"
"sort"
"strconv"
"strings"
"git.noroutes.net/NoRoutes/vibetunes/internal/database"
"git.noroutes.net/NoRoutes/vibetunes/internal/similarity"
"git.noroutes.net/NoRoutes/vibetunes/internal/streaming"
)
@@ -182,18 +184,32 @@ func handleGetSimilarSongs2(rc *Context) *apiError {
}
// similarSongsFor resolves id as a song, album, or artist (getSimilarSongs's
// "id" can be any of the three in real Subsonic usage) and finds
// genre-similar songs for it.
// "id" can be any of the three in real Subsonic usage), finds
// genre-similar songs for it, and — when the seed song has been sonic-
// analyzed — re-ranks that genre-matched candidate set by actual sonic
// distance rather than leaving it in the database's random order. This
// never replaces the genre filter with pure sonic search (that's what
// getSonicSimilarTracks is for); it just makes "similar" among
// genre-mates mean something more concrete when the data exists to
// back it up.
func similarSongsFor(db *sql.DB, id string, count int) ([]*database.Song, error) {
if song, err := database.GetSong(db, id); err == nil {
return database.SimilarSongsByGenre(db, song.ID, count)
songs, err := database.SimilarSongsByGenre(db, song.ID, count)
if err != nil {
return nil, err
}
return reorderBySonicSimilarity(db, song.ID, songs), nil
}
if _, err := database.GetAlbum(db, id); err == nil {
songs, err := database.ListSongsByAlbum(db, id)
if err != nil || len(songs) == 0 {
return nil, err
}
return database.SimilarSongsByGenre(db, songs[0].ID, count)
similar, err := database.SimilarSongsByGenre(db, songs[0].ID, count)
if err != nil {
return nil, err
}
return reorderBySonicSimilarity(db, songs[0].ID, similar), nil
}
if _, err := database.GetArtist(db, id); err == nil {
return database.SimilarSongsByArtist(db, id, count)
@@ -201,6 +217,50 @@ func similarSongsFor(db *sql.DB, id string, count int) ([]*database.Song, error)
return nil, database.ErrNotFound
}
// reorderBySonicSimilarity sorts songs by real sonic distance from
// seedSongID when both the seed and (some of) the candidates have
// stored audio_features; candidates without analysis data are left in
// their existing relative order at the end, never dropped and never
// assigned a fabricated distance.
func reorderBySonicSimilarity(db *sql.DB, seedSongID string, songs []*database.Song) []*database.Song {
if len(songs) < 2 {
return songs
}
seed, err := database.GetAudioFeatures(db, seedSongID)
if err != nil || seed == nil {
return songs
}
items := []similarity.Item{{ID: seedSongID, Vector: seed.Vector()}}
analyzed := map[string]bool{}
for _, s := range songs {
f, err := database.GetAudioFeatures(db, s.ID)
if err != nil || f == nil {
continue
}
items = append(items, similarity.Item{ID: s.ID, Vector: f.Vector()})
analyzed[s.ID] = true
}
if len(items) < 2 {
return songs
}
idx := similarity.Build(items)
sorted := make([]*database.Song, len(songs))
copy(sorted, songs)
sort.SliceStable(sorted, func(i, j int) bool {
ai, aj := analyzed[sorted[i].ID], analyzed[sorted[j].ID]
if ai != aj {
return ai // analyzed songs sort before unanalyzed ones
}
if !ai {
return false // preserve relative order among unanalyzed songs
}
return idx.DistanceBetween(seedSongID, sorted[i].ID) < idx.DistanceBetween(seedSongID, sorted[j].ID)
})
return sorted
}
func handleGetTopSongs(rc *Context) *apiError {
count := rc.Params.Int("count", 50)
db := rc.Server.DB