feat(transcode): add a generic transcoding package for encoding/decoding/caching

2022-03-10 00:42:52 +00:00
parent fd211d706a
commit 165904c2bb
25 changed files with 713 additions and 388 deletions
--- a/go.mod
+++ b/go.mod
@@ -11,6 +11,7 @@ require (
 	github.com/disintegration/imaging v1.6.2
 	github.com/dustin/go-humanize v1.0.0
 	github.com/faiface/beep v1.1.0
+	github.com/google/shlex v0.0.0-20191202100458-e7afc7fbc510
 	github.com/google/uuid v1.3.0 // indirect
 	github.com/gorilla/mux v1.8.0
 	github.com/gorilla/securecookie v1.1.1
@@ -26,7 +27,7 @@ require (
 	github.com/json-iterator/go v1.1.12 // indirect
 	github.com/lib/pq v1.3.0 // indirect
 	github.com/matryer/is v1.4.0
-	github.com/mattn/go-sqlite3 v1.14.11 // indirect
+	github.com/mattn/go-sqlite3 v1.14.11
 	github.com/mewkiz/pkg v0.0.0-20211102230744-16a6ce8f1b77 // indirect
 	github.com/mitchellh/copystructure v1.2.0 // indirect
 	github.com/mmcdole/gofeed v1.1.3
@@ -40,7 +41,7 @@ require (
 	github.com/sentriz/gormstore v0.0.0-20220105134332-64e31f7f6981
 	github.com/stretchr/testify v1.7.0 // indirect
 	golang.org/x/crypto v0.0.0-20220209155544-dad33157f4bf // indirect
-	golang.org/x/exp/shiny v0.0.0-20220209042442-160e291fcf24 // indirect
+	golang.org/x/exp/shiny v0.0.0-20220407100705-7b9b53b0aca4 // indirect
 	golang.org/x/image v0.0.0-20211028202545-6944b10bf410 // indirect
 	golang.org/x/mobile v0.0.0-20220112015953-858099ff7816 // indirect
 	golang.org/x/net v0.0.0-20220127200216-cd36cc0744dd // indirect
--- a/go.sum
+++ b/go.sum
@@ -22,7 +22,6 @@ github.com/cespare/xxhash v1.1.0/go.mod h1:XrSqR1VqqWfGrhpAt58auRo0WTKS1nRRg3ghf
 github.com/cpuguy83/go-md2man/v2 v2.0.0-20190314233015-f79a8a8ca69d/go.mod h1:maD7wRr/U5Z6m/iR4s+kqSMx2CaBsrgA7czyZG/E6dU=
 github.com/d4l3k/messagediff v1.2.2-0.20190829033028-7e0a312ae40b/go.mod h1:Oozbb1TVXFac9FtSIxHBMnBCq2qeH/2KkEQxENCrlLo=
 github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
-github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c=
 github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
 github.com/denisenkom/go-mssqldb v0.0.0-20181014144952-4e0d7dc8888f/go.mod h1:xN/JuLBIz4bjkxNmByTiV1IbhfnYb6oo99phBn4Eqhc=
 github.com/denisenkom/go-mssqldb v0.0.0-20191124224453-732737034ffd h1:83Wprp6ROGeiHFAP8WJdI2RoxALQYgdllERc3N5N2DM=
@@ -50,6 +49,8 @@ github.com/golang-sql/civil v0.0.0-20190719163853-cb61b32ac6fe/go.mod h1:8vg3r2V
 github.com/golang/protobuf v1.2.0/go.mod h1:6lQm79b+lXiMfvg/cZm0SGofjICqVBUtrP5yJMmIC1U=
 github.com/google/go-cmp v0.2.0/go.mod h1:oXzfMopK8JAjlY9xF4vHSVASa0yLyX7SntLO5aqRK0M=
 github.com/google/gofuzz v1.0.0/go.mod h1:dBl0BpW6vV/+mYPU4Po3pmUjxk6FQPldtuIdl/M65Eg=
+github.com/google/shlex v0.0.0-20191202100458-e7afc7fbc510 h1:El6M4kTTCOh6aBiKaUGG7oYTSPP8MxqL4YI3kZKwcP4=
+github.com/google/shlex v0.0.0-20191202100458-e7afc7fbc510/go.mod h1:pupxD2MaaD3pAXIBCelhxNneeOaAeabZDe5s4K6zSpQ=
 github.com/google/uuid v1.3.0 h1:t6JiXgmwXMjEs8VusXIJk2BXHsn+wx8BZdTaoZ5fu7I=
 github.com/google/uuid v1.3.0/go.mod h1:TIyPZe4MgqvfeYDBFedMoGGpEw/LqOeaOT+nhxU+yHo=
 github.com/gorilla/context v1.1.1 h1:AWwleXJkX/nhcU9bZSnZoi3h/qGYqQAGhq6zZe/aQW8=
@@ -76,6 +77,7 @@ github.com/icza/mighty v0.0.0-20180919140131-cfd07d671de6 h1:8UsGZ2rr2ksmEru6lTo
 github.com/icza/mighty v0.0.0-20180919140131-cfd07d671de6/go.mod h1:xQig96I1VNBDIWGCdTt54nHt6EeI639SmHycLYL7FkA=
 github.com/imdario/mergo v0.3.12 h1:b6R2BslTbIEToALKP7LxUvijTsNI9TAe80pLWN2g/HU=
 github.com/imdario/mergo v0.3.12/go.mod h1:jmQim1M+e3UYxmgPu/WyfjB3N3VflVyUjjjwH0dnCYA=
+github.com/jezek/xgb v1.0.0/go.mod h1:nrhwO0FX/enq75I7Y7G8iN1ubpSGZEiA3v9e9GyRFlk=
 github.com/jfreymuth/oggvorbis v1.0.1/go.mod h1:NqS+K+UXKje0FUYUPosyQ+XTVvjmVjps1aEZH1sumIk=
 github.com/jfreymuth/vorbis v1.0.0/go.mod h1:8zy3lUAm9K/rJJk223RKy6vjCZTWC61NA2QD06bfOE0=
 github.com/jinzhu/gorm v1.9.2/go.mod h1:Vla75njaFJ8clLU1W44h34PjIkijhjHIYnZxMqCdxqo=
@@ -170,6 +172,8 @@ golang.org/x/exp v0.0.0-20190731235908-ec7cb31e5a56 h1:estk1glOnSVeJ9tdEZZc5mAMD
 golang.org/x/exp v0.0.0-20190731235908-ec7cb31e5a56/go.mod h1:JhuoJpWY28nO4Vef9tZUw9qufEGTyX1+7lmHxV5q5G4=
 golang.org/x/exp/shiny v0.0.0-20220209042442-160e291fcf24 h1:jn6Q9FOmCn1Kk7ec3Qm9lfygAr7dv8J1YfEx6RQcRJQ=
 golang.org/x/exp/shiny v0.0.0-20220209042442-160e291fcf24/go.mod h1:NtXcNtv5Wu0zUbBl574y/D5MMZvnQnV3sgjZxbs64Jo=
+golang.org/x/exp/shiny v0.0.0-20220407100705-7b9b53b0aca4 h1:ywNGLBFk8tKaiu+GYZeoXWzrFoJ/a1LHYKy1lb3R9cM=
+golang.org/x/exp/shiny v0.0.0-20220407100705-7b9b53b0aca4/go.mod h1:VjAR7z0ngyATZTELrBSkxOOHhhlnVUxDye4mcjx5h/8=
 golang.org/x/image v0.0.0-20190220214146-31aff87c08e9/go.mod h1:kZ7UVZpmo3dzQBMxlp+ypCbDeSB+sBbTgSJuh5dn5js=
 golang.org/x/image v0.0.0-20190227222117-0694c2d4d067/go.mod h1:kZ7UVZpmo3dzQBMxlp+ypCbDeSB+sBbTgSJuh5dn5js=
 golang.org/x/image v0.0.0-20190802002840-cff245a6509b/go.mod h1:FeLwcggjj3mMvU+oOTbSwawSJRM1uh48EjtB4UJZlP0=
--- a/server/ctrladmin/handlers.go
+++ b/server/ctrladmin/handlers.go
@@ -10,10 +10,10 @@ import (
 	"github.com/mmcdole/gofeed"

 	"go.senan.xyz/gonic/server/db"
-	"go.senan.xyz/gonic/server/encode"
 	"go.senan.xyz/gonic/server/scanner"
 	"go.senan.xyz/gonic/server/scrobble/lastfm"
 	"go.senan.xyz/gonic/server/scrobble/listenbrainz"
+	"go.senan.xyz/gonic/server/transcode"
 )

 func doScan(scanner *scanner.Scanner, opts scanner.ScanOptions) {
@@ -67,7 +67,7 @@ func (c *Controller) ServeHome(r *http.Request) *Response {
 	c.DB.
 		Where("user_id=?", user.ID).
 		Find(&data.TranscodePreferences)
-	for profile := range encode.Profiles() {
+	for profile := range transcode.UserProfiles {
 		data.TranscodeProfiles = append(data.TranscodeProfiles, profile)
 	}
 	// podcasts box
--- a/server/ctrlsubsonic/ctrl.go
+++ b/server/ctrlsubsonic/ctrl.go
@@ -15,6 +15,7 @@ import (
 	"go.senan.xyz/gonic/server/jukebox"
 	"go.senan.xyz/gonic/server/podcasts"
 	"go.senan.xyz/gonic/server/scrobble"
+	"go.senan.xyz/gonic/server/transcode"
 )

 type CtxKey int
@@ -34,6 +35,7 @@ type Controller struct {
 	Jukebox        *jukebox.Jukebox
 	Scrobblers     []scrobble.Scrobbler
 	Podcasts       *podcasts.Podcasts
+	Transcoder     transcode.Transcoder
 }

 type metaResponse struct {
--- a/server/ctrlsubsonic/ctrl_test.go
+++ b/server/ctrlsubsonic/ctrl_test.go
@@ -18,12 +18,22 @@ import (

 	"go.senan.xyz/gonic/server/ctrlbase"
 	"go.senan.xyz/gonic/server/ctrlsubsonic/params"
+	"go.senan.xyz/gonic/server/db"
 	"go.senan.xyz/gonic/server/mockfs"
+	"go.senan.xyz/gonic/server/transcode"
 )

-var (
-	testDataDir   = "testdata"
-	testCamelExpr = regexp.MustCompile("([a-z0-9])([A-Z])")
+var testCamelExpr = regexp.MustCompile("([a-z0-9])([A-Z])")
+
+const (
+	mockUsername   = "admin"
+	mockPassword   = "admin"
+	mockClientName = "test"
+)
+
+const (
+	audioPath5s  = "testdata/audio/5s.flac"  //nolint:deadcode,varcheck
+	audioPath10s = "testdata/audio/10s.flac" //nolint:deadcode,varcheck
 )

 type queryCase struct {
@@ -37,22 +47,43 @@ func makeGoldenPath(test string) string {
 	snake := testCamelExpr.ReplaceAllString(test, "${1}_${2}")
 	lower := strings.ToLower(snake)
 	relPath := strings.ReplaceAll(lower, "/", "_")
-	return path.Join(testDataDir, relPath)
+	return path.Join("testdata", relPath)
 }

 func makeHTTPMock(query url.Values) (*httptest.ResponseRecorder, *http.Request) {
 	// ensure the handlers give us json
 	query.Add("f", "json")
+	query.Add("u", mockUsername)
+	query.Add("p", mockPassword)
+	query.Add("v", "1")
+	query.Add("c", mockClientName)
 	// request from the handler in question
 	req, _ := http.NewRequest("", "", nil)
 	req.URL.RawQuery = query.Encode()
-	subParams := params.New(req)
-	withParams := context.WithValue(req.Context(), CtxParams, subParams)
+	ctx := req.Context()
+	ctx = context.WithValue(ctx, CtxParams, params.New(req))
+	ctx = context.WithValue(ctx, CtxUser, &db.User{})
+	req = req.WithContext(ctx)
 	rr := httptest.NewRecorder()
-	req = req.WithContext(withParams)
 	return rr, req
 }

+func serveRaw(t *testing.T, contr *Controller, h handlerSubsonicRaw, rr *httptest.ResponseRecorder, req *http.Request) {
+	type middleware func(http.Handler) http.Handler
+	middlewares := []middleware{
+		contr.WithParams,
+		contr.WithRequiredParams,
+		contr.WithUser,
+	}
+
+	handler := contr.HR(h)
+	for _, m := range middlewares {
+		handler = m(handler)
+	}
+
+	handler.ServeHTTP(rr, req)
+}
+
 func runQueryCases(t *testing.T, contr *Controller, h handlerSubsonic, cases []*queryCase) {
 	t.Helper()
 	for _, qc := range cases {
@@ -96,20 +127,26 @@ func runQueryCases(t *testing.T, contr *Controller, h handlerSubsonic, cases []*
 	}
 }

-func makeController(t *testing.T) *Controller                  { return makec(t, []string{""}) }
-func makeControllerRoots(t *testing.T, r []string) *Controller { return makec(t, r) }
+func makeController(t *testing.T) *Controller                  { return makec(t, []string{""}, false) }
+func makeControllerRoots(t *testing.T, r []string) *Controller { return makec(t, r, false) }
+func makeControllerAudio(t *testing.T) *Controller             { return makec(t, []string{""}, true) }

-func makec(t *testing.T, roots []string) *Controller {
+func makec(t *testing.T, roots []string, audio bool) *Controller {
 	t.Helper()

 	m := mockfs.NewWithDirs(t, roots)
 	for _, root := range roots {
 		m.AddItemsPrefixWithCovers(root)
+		if !audio {
+			continue
+		}
+		m.SetRealAudio(filepath.Join(root, "artist-0/album-0/track-0.flac"), 10, audioPath10s)
+		m.SetRealAudio(filepath.Join(root, "artist-0/album-0/track-1.flac"), 10, audioPath10s)
+		m.SetRealAudio(filepath.Join(root, "artist-0/album-0/track-2.flac"), 10, audioPath10s)
 	}

 	m.ScanAndClean()
 	m.ResetDates()
-	m.LogAlbums()

 	var absRoots []string
 	for _, root := range roots {
@@ -117,7 +154,13 @@ func makec(t *testing.T, roots []string) *Controller {
 	}

 	base := &ctrlbase.Controller{DB: m.DB()}
-	return &Controller{Controller: base, MusicPaths: absRoots}
+	contr := &Controller{
+		Controller: base,
+		MusicPaths: absRoots,
+		Transcoder: transcode.NewFFmpegTranscoder(),
+	}
+
+	return contr
 }

 func TestMain(m *testing.M) {
--- a/server/ctrlsubsonic/handlers_raw.go
+++ b/server/ctrlsubsonic/handlers_raw.go
@@ -3,7 +3,6 @@ package ctrlsubsonic
 import (
 	"errors"
 	"fmt"
-	"io"
 	"log"
 	"net/http"
 	"os"
@@ -13,12 +12,13 @@ import (
 	"github.com/disintegration/imaging"
 	"github.com/jinzhu/gorm"

+	"go.senan.xyz/gonic/iout"
+	"go.senan.xyz/gonic/server/ctrlsubsonic/httprange"
 	"go.senan.xyz/gonic/server/ctrlsubsonic/params"
 	"go.senan.xyz/gonic/server/ctrlsubsonic/spec"
 	"go.senan.xyz/gonic/server/ctrlsubsonic/specid"
 	"go.senan.xyz/gonic/server/db"
-	"go.senan.xyz/gonic/server/encode"
-	"go.senan.xyz/gonic/server/mime"
+	"go.senan.xyz/gonic/server/transcode"
 )

 // "raw" handlers are ones that don't always return a spec response.
@@ -36,7 +36,7 @@ func streamGetTransPref(dbc *db.DB, userID int, client string) (*db.TranscodePre
 		First(&pref).
 		Error
 	if errors.Is(err, gorm.ErrRecordNotFound) {
-		return &pref, nil
+		return nil, nil
 	}
 	if err != nil {
 		return nil, fmt.Errorf("find transcode preference: %w", err)
@@ -242,7 +242,7 @@ func (c *Controller) ServeStream(w http.ResponseWriter, r *http.Request) *spec.R
 	if err != nil {
 		return spec.NewError(10, "please provide an `id` parameter")
 	}
-	var audioFile db.AudioFile
+	var file db.AudioFile
 	var audioPath string
 	switch id.Type {
 	case specid.Track:
@@ -250,103 +250,78 @@ func (c *Controller) ServeStream(w http.ResponseWriter, r *http.Request) *spec.R
 		if err != nil {
 			return spec.NewError(70, "track with id `%s` was not found", id)
 		}
-		audioFile = track
+		file = track
 		audioPath = path.Join(track.AbsPath())
 	case specid.PodcastEpisode:
 		podcast, err := streamGetPodcast(c.DB, id.Value)
 		if err != nil {
 			return spec.NewError(70, "podcast with id `%s` was not found", id)
 		}
-		audioFile = podcast
+		file = podcast
 		audioPath = path.Join(c.PodcastsPath, podcast.Path)
 	default:
 		return spec.NewError(70, "media type of `%s` was not found", id.Type)
 	}

 	user := r.Context().Value(CtxUser).(*db.User)
-	if track, ok := audioFile.(*db.Track); ok && track.Album != nil {
+	if track, ok := file.(*db.Track); ok && track.Album != nil {
 		defer func() {
 			if err := streamUpdateStats(c.DB, user.ID, track.Album.ID, time.Now()); err != nil {
-				log.Printf("error updating listen stats: %v", err)
+				log.Printf("error updating status: %v", err)
 			}
 		}()
 	}

-	pref, err := streamGetTransPref(c.DB, user.ID, params.GetOr("c", ""))
-	if err != nil {
-		return spec.NewError(0, "failed to get transcode stream preference: %v", err)
-	}
-
-	onInvalidProfile := func() error {
-		log.Printf("serving raw `%s`\n", audioFile.AudioFilename())
-		w.Header().Set("Content-Type", audioFile.MIME())
+	if format, _ := params.Get("format"); format == "raw" {
 		http.ServeFile(w, r, audioPath)
 		return nil
 	}
-	onCacheHit := func(profile encode.Profile, path string) error {
-		log.Printf("serving transcode `%s`: cache [%s/%dk] hit!\n",
-			audioFile.AudioFilename(), profile.Format, profile.Bitrate)
-		cacheMime, _ := mime.FromExtension(profile.Format)
-		w.Header().Set("Content-Type", cacheMime)

-		cacheFile, err := os.Stat(path)
-		if err != nil {
-			return fmt.Errorf("failed to stat cache file `%s`: %w", path, err)
-		}
-		contentLength := fmt.Sprintf("%d", cacheFile.Size())
-		w.Header().Set("Content-Length", contentLength)
-		http.ServeFile(w, r, path)
+	pref, err := streamGetTransPref(c.DB, user.ID, params.GetOr("c", ""))
+	if err != nil && !errors.Is(err, gorm.ErrRecordNotFound) {
+		return spec.NewError(0, "couldn't find transcode preference: %v", err)
+	}
+	if pref == nil {
+		http.ServeFile(w, r, audioPath)
 		return nil
 	}
-	onCacheMiss := func(profile encode.Profile) (io.Writer, error) {
-		log.Printf("serving transcode `%s`: cache [%s/%dk] miss!\n",
-			audioFile.AudioFilename(), profile.Format, profile.Bitrate)
-		encodeMime, _ := mime.FromExtension(profile.Format)
-		w.Header().Set("Content-Type", encodeMime)
-		return w, nil
-	}
-	encodeOptions := encode.Options{
-		TrackPath:        audioPath,
-		TrackBitrate:     audioFile.AudioBitrate(),
-		CachePath:        c.CachePath,
-		ProfileName:      pref.Profile,
-		PreferredBitrate: params.GetOrInt("maxBitRate", 0),
-		OnInvalidProfile: onInvalidProfile,
-		OnCacheHit:       onCacheHit,
-		OnCacheMiss:      onCacheMiss,
-	}
-	if err := encode.Encode(encodeOptions); err != nil {
-		log.Printf("serving transcode `%s`: error: %v\n", audioFile.AudioFilename(), err)
-	}
-	return nil
-}

-func (c *Controller) ServeDownload(w http.ResponseWriter, r *http.Request) *spec.Response {
-	params := r.Context().Value(CtxParams).(params.Params)
-	id, err := params.GetID("id")
+	profile, ok := transcode.UserProfiles[pref.Profile]
+	if !ok {
+		return spec.NewError(0, "unknown transcode user profile %q", pref.Profile)
+	}
+	if max, _ := params.GetInt("maxBitRate"); max > 0 && int(profile.BitRate()) > max {
+		profile = transcode.WithBitrate(profile, transcode.BitRate(max))
+	}
+
+	log.Printf("trancoding to %q with max bitrate %dk", profile.MIME(), profile.BitRate())
+
+	transcodeReader, err := c.Transcoder.Transcode(r.Context(), profile, audioPath)
 	if err != nil {
-		return spec.NewError(10, "please provide an `id` parameter")
+		return spec.NewError(0, "error transcoding: %v", err)
 	}
-	var filePath string
-	var audioFile db.AudioFile
-	switch id.Type {
-	case specid.Track:
-		track, _ := streamGetTrack(c.DB, id.Value)
-		audioFile = track
-		filePath = track.AbsPath()
-		if err != nil {
-			return spec.NewError(70, "track with id `%s` was not found", id)
-		}
-	case specid.PodcastEpisode:
-		podcast, err := streamGetPodcast(c.DB, id.Value)
-		audioFile = podcast
-		filePath = path.Join(c.PodcastsPath, podcast.Path)
-		if err != nil {
-			return spec.NewError(70, "podcast with id `%s` was not found", id)
-		}
+	defer transcodeReader.Close()
+
+	length := transcode.GuessExpectedSize(profile, time.Duration(file.AudioLength())*time.Second) // TODO: if there's no duration?
+	rreq, err := httprange.Parse(r.Header.Get("Range"), length)
+	if err != nil {
+		return spec.NewError(0, "error parsing range: %v", err)
+	}
+
+	w.Header().Set("Content-Type", profile.MIME())
+	w.Header().Set("Content-Length", fmt.Sprintf("%d", rreq.Length))
+	w.Header().Set("Accept-Ranges", string(httprange.UnitBytes))
+
+	if rreq.Partial {
+		w.WriteHeader(http.StatusPartialContent)
+		w.Header().Set("Content-Range", fmt.Sprintf("%s %d-%d/%d", httprange.UnitBytes, rreq.Start, rreq.End, length))
+	}
+
+	if err := iout.CopyRange(w, transcodeReader, int64(rreq.Start), int64(rreq.Length)); err != nil {
+		log.Printf("error writing transcoded data: %v", err)
+	}
+	if f, ok := w.(http.Flusher); ok {
+		f.Flush()
 	}
-	log.Printf("serving raw `%s`\n", audioFile.AudioFilename())
-	w.Header().Set("Content-Type", audioFile.MIME())
-	http.ServeFile(w, r, filePath)
 	return nil
 }
--- a/server/ctrlsubsonic/handlers_raw_test.go
+++ b/server/ctrlsubsonic/handlers_raw_test.go
@@ -0,0 +1,152 @@
+package ctrlsubsonic
+
+import (
+	"fmt"
+	"io/fs"
+	"net/http"
+	"net/url"
+	"os"
+	"os/exec"
+	"strconv"
+	"testing"
+	"time"
+
+	"github.com/matryer/is"
+	"go.senan.xyz/gonic/server/db"
+	"go.senan.xyz/gonic/server/transcode"
+)
+
+func TestServeStreamRaw(t *testing.T) {
+	t.Parallel()
+	if _, err := exec.LookPath("ffmpeg"); err != nil {
+		t.Skipf("no ffmpeg in $PATH")
+	}
+
+	is := is.New(t)
+	contr := makeControllerAudio(t)
+
+	statFlac := stat(t, audioPath10s)
+
+	rr, req := makeHTTPMock(url.Values{"id": {"tr-1"}})
+	serveRaw(t, contr, contr.ServeStream, rr, req)
+
+	is.Equal(rr.Code, http.StatusOK)
+	is.Equal(rr.Header().Get("content-type"), "audio/flac")
+	is.Equal(atoi(t, rr.Header().Get("content-length")), int(statFlac.Size()))
+	is.Equal(atoi(t, rr.Header().Get("content-length")), rr.Body.Len())
+}
+
+func TestServeStreamOpus(t *testing.T) {
+	t.Parallel()
+	if _, err := exec.LookPath("ffmpeg"); err != nil {
+		t.Skipf("no ffmpeg in $PATH")
+	}
+
+	is := is.New(t)
+	contr := makeControllerAudio(t)
+
+	var user db.User
+	is.NoErr(contr.DB.Where("name=?", mockUsername).Find(&user).Error)
+	is.NoErr(contr.DB.Create(&db.TranscodePreference{UserID: user.ID, Client: mockClientName, Profile: "opus"}).Error)
+
+	rr, req := makeHTTPMock(url.Values{"id": {"tr-1"}})
+	serveRaw(t, contr, contr.ServeStream, rr, req)
+
+	is.Equal(rr.Code, http.StatusOK)
+	is.Equal(rr.Header().Get("content-type"), "audio/ogg")
+	is.Equal(atoi(t, rr.Header().Get("content-length")), transcode.GuessExpectedSize(transcode.Opus, 10*time.Second))
+	is.Equal(atoi(t, rr.Header().Get("content-length")), rr.Body.Len())
+}
+
+func TestServeStreamOpusMaxBitrate(t *testing.T) {
+	t.Parallel()
+	if _, err := exec.LookPath("ffmpeg"); err != nil {
+		t.Skipf("no ffmpeg in $PATH")
+	}
+
+	is := is.New(t)
+	contr := makeControllerAudio(t)
+
+	var user db.User
+	is.NoErr(contr.DB.Where("name=?", mockUsername).Find(&user).Error)
+	is.NoErr(contr.DB.Create(&db.TranscodePreference{UserID: user.ID, Client: mockClientName, Profile: "opus"}).Error)
+
+	const bitrate = 5
+
+	rr, req := makeHTTPMock(url.Values{"id": {"tr-1"}, "maxBitRate": {strconv.Itoa(bitrate)}})
+	serveRaw(t, contr, contr.ServeStream, rr, req)
+
+	profile := transcode.WithBitrate(transcode.Opus, transcode.BitRate(bitrate))
+	expectedLength := transcode.GuessExpectedSize(profile, 10*time.Second)
+
+	is.Equal(rr.Code, http.StatusOK)
+	is.Equal(rr.Header().Get("content-type"), "audio/ogg")
+	is.Equal(atoi(t, rr.Header().Get("content-length")), expectedLength)
+	is.Equal(atoi(t, rr.Header().Get("content-length")), rr.Body.Len())
+}
+
+func TestServeStreamMP3Range(t *testing.T) {
+	t.Parallel()
+	if _, err := exec.LookPath("ffmpeg"); err != nil {
+		t.Skipf("no ffmpeg in $PATH")
+	}
+
+	is := is.New(t)
+	contr := makeControllerAudio(t)
+
+	var user db.User
+	is.NoErr(contr.DB.Where("name=?", mockUsername).Find(&user).Error)
+	is.NoErr(contr.DB.Create(&db.TranscodePreference{UserID: user.ID, Client: mockClientName, Profile: "mp3"}).Error)
+
+	var totalBytes []byte
+	{
+		rr, req := makeHTTPMock(url.Values{"id": {"tr-1"}})
+		serveRaw(t, contr, contr.ServeStream, rr, req)
+		is.Equal(rr.Code, http.StatusOK)
+		is.Equal(rr.Header().Get("content-type"), "audio/mpeg")
+		totalBytes = rr.Body.Bytes()
+	}
+
+	const chunkSize = 2 << 16
+
+	var bytes []byte
+	for i := 0; i < len(totalBytes); i += chunkSize {
+		rr, req := makeHTTPMock(url.Values{"id": {"tr-1"}})
+		req.Header.Set("range", fmt.Sprintf("bytes=%d-%d", i, min(i+chunkSize, len(totalBytes))-1))
+		t.Log(req.Header.Get("range"))
+		serveRaw(t, contr, contr.ServeStream, rr, req)
+		is.Equal(rr.Code, http.StatusPartialContent)
+		is.Equal(rr.Header().Get("content-type"), "audio/mpeg")
+		is.True(atoi(t, rr.Header().Get("content-length")) == chunkSize || atoi(t, rr.Header().Get("content-length")) == len(totalBytes)%chunkSize)
+		is.Equal(atoi(t, rr.Header().Get("content-length")), rr.Body.Len())
+		bytes = append(bytes, rr.Body.Bytes()...)
+	}
+
+	is.Equal(len(totalBytes), len(bytes))
+	is.Equal(totalBytes, bytes)
+}
+
+func stat(t *testing.T, path string) fs.FileInfo {
+	t.Helper()
+	info, err := os.Stat(path)
+	if err != nil {
+		t.Fatalf("stat %q: %v", path, err)
+	}
+	return info
+}
+
+func atoi(t *testing.T, in string) int {
+	t.Helper()
+	i, err := strconv.Atoi(in)
+	if err != nil {
+		t.Fatalf("atoi %q: %v", in, err)
+	}
+	return i
+}
+
+func min(a, b int) int {
+	if a < b {
+		return a
+	}
+	return b
+}
--- a/server/ctrlsubsonic/httprange/httprange.go
+++ b/server/ctrlsubsonic/httprange/httprange.go
@@ -0,0 +1,59 @@
+package httprange
+
+import (
+	"fmt"
+	"regexp"
+	"strconv"
+)
+
+type Unit string
+
+const (
+	UnitBytes Unit = "bytes"
+)
+
+//nolint:gochecknoglobals
+var (
+	reg   = regexp.MustCompile(`^(?P<unit>\w+)=(?P<start>(?:\d+)?)\s*-\s*(?P<end>(?:\d+)?)$`)
+	unit  = reg.SubexpIndex("unit")
+	start = reg.SubexpIndex("start")
+	end   = reg.SubexpIndex("end")
+)
+
+var (
+	ErrInvalidRange = fmt.Errorf("invalid range")
+	ErrUnknownUnit  = fmt.Errorf("unknown range")
+)
+
+type Range struct {
+	Start, End, Length int // bytes
+	Partial            bool
+}
+
+func Parse(in string, fullLength int) (Range, error) {
+	parts := reg.FindStringSubmatch(in)
+	if len(parts)-1 != reg.NumSubexp() {
+		return Range{0, fullLength - 1, fullLength, false}, nil
+	}
+
+	switch unit := parts[unit]; Unit(unit) {
+	case UnitBytes:
+	default:
+		return Range{}, fmt.Errorf("%q: %w", unit, ErrUnknownUnit)
+	}
+
+	start, _ := strconv.Atoi(parts[start])
+	end, _ := strconv.Atoi(parts[end])
+	length := fullLength
+	partial := false
+
+	switch {
+	case end > 0 && end < length:
+		length = end - start + 1
+		partial = true
+	case end == 0 && length > 0:
+		end = length - 1
+	}
+
+	return Range{start, end, length, partial}, nil
+}
--- a/server/ctrlsubsonic/httprange/httprange_test.go
+++ b/server/ctrlsubsonic/httprange/httprange_test.go
@@ -0,0 +1,30 @@
+package httprange_test
+
+import (
+	"testing"
+
+	"github.com/matryer/is"
+	"go.senan.xyz/gonic/server/ctrlsubsonic/httprange"
+)
+
+func TestParse(t *testing.T) {
+	is := is.New(t)
+
+	full := func(start, end, length int) httprange.Range {
+		return httprange.Range{Start: start, End: end, Length: length}
+	}
+	partial := func(start, end, length int) httprange.Range {
+		return httprange.Range{Start: start, End: end, Length: length, Partial: true}
+	}
+	parse := func(in string, length int) httprange.Range {
+		is.Helper()
+		rrange, err := httprange.Parse(in, length)
+		is.NoErr(err)
+		return rrange
+	}
+
+	is.Equal(parse("bytes=0-0", 0), full(0, 0, 0))
+	is.Equal(parse("bytes=0-", 10), full(0, 9, 10))
+	is.Equal(parse("bytes=0-49", 50), partial(0, 49, 50))
+	is.Equal(parse("bytes=50-99", 100), partial(50, 99, 50))
+}
--- a/server/ctrlsubsonic/testdata/audio/10s.flac
+++ b/server/ctrlsubsonic/testdata/audio/10s.flac
--- a/server/ctrlsubsonic/testdata/audio/5s.flac
+++ b/server/ctrlsubsonic/testdata/audio/5s.flac
--- a/server/db/model.go
+++ b/server/db/model.go
@@ -72,10 +72,11 @@ type Genre struct {
 // AudioFile is used to avoid some duplication in handlers_raw.go
 // between Track and Podcast
 type AudioFile interface {
-	AudioFilename() string
 	Ext() string
 	MIME() string
+	AudioFilename() string
 	AudioBitrate() int
+	AudioLength() int
 }

 type Track struct {
@@ -100,6 +101,9 @@ type Track struct {
 	TagBrainzID    string   `sql:"default: null"`
 }

+func (t *Track) AudioLength() int  { return t.Length }
+func (t *Track) AudioBitrate() int { return t.Bitrate }
+
 func (t *Track) SID() *specid.ID {
 	return &specid.ID{Type: specid.Track, Value: t.ID}
 }
@@ -124,10 +128,6 @@ func (t *Track) AudioFilename() string {
 	return t.Filename
 }

-func (t *Track) AudioBitrate() int {
-	return t.Bitrate
-}
-
 func (t *Track) MIME() string {
 	v, _ := mime.FromExtension(t.Ext())
 	return v
@@ -364,6 +364,9 @@ type PodcastEpisode struct {
 	Error       string
 }

+func (pe *PodcastEpisode) AudioLength() int  { return pe.Length }
+func (pe *PodcastEpisode) AudioBitrate() int { return pe.Bitrate }
+
 func (pe *PodcastEpisode) SID() *specid.ID {
 	return &specid.ID{Type: specid.PodcastEpisode, Value: pe.ID}
 }
@@ -385,10 +388,6 @@ func (pe *PodcastEpisode) MIME() string {
 	return v
 }

-func (pe *PodcastEpisode) AudioBitrate() int {
-	return pe.Bitrate
-}
-
 type Bookmark struct {
 	ID          int `gorm:"primary_key"`
 	User        *User
--- a/server/encode/encode.go
+++ b/server/encode/encode.go
@@ -1,284 +0,0 @@
-// author: spijet (https://github.com/spijet/)
-
-package encode
-
-import (
-	"fmt"
-	"io"
-	"log"
-	"net/http"
-	"os"
-	"os/exec"
-	"path"
-	"strings"
-
-	"github.com/cespare/xxhash"
-)
-
-const (
-	buffLen = 4096
-	ffmpeg  = "ffmpeg"
-)
-
-type replayGain int
-
-const (
-	rgForce replayGain = iota
-	rgHigh
-)
-
-type Profile struct {
-	Format  string
-	Bitrate int
-
-	options    []string
-	replayGain replayGain
-	upsample   bool
-}
-
-func fileExists(filename string) bool {
-	info, err := os.Stat(filename)
-	if os.IsNotExist(err) {
-		return false
-	}
-	return !info.IsDir()
-}
-
-func Profiles() map[string]Profile {
-	return map[string]Profile{
-		"mp3": {
-			Format:  "mp3",
-			Bitrate: 128,
-			options: []string{"-c:a", "libmp3lame"},
-		},
-		"mp3_rg": {
-			Format:     "mp3",
-			Bitrate:    128,
-			options:    []string{"-c:a", "libmp3lame"},
-			replayGain: rgForce,
-		},
-		"opus": {
-			Format:  "opus",
-			Bitrate: 96,
-			options: []string{"-c:a", "libopus", "-vbr", "on"},
-		},
-		"opus_rg": {
-			Format:     "opus",
-			Bitrate:    96,
-			options:    []string{"-c:a", "libopus", "-vbr", "on"},
-			replayGain: rgForce,
-		},
-		"opus_car": {
-			Format:     "opus",
-			Bitrate:    96,
-			options:    []string{"-c:a", "libopus", "-vbr", "on"},
-			replayGain: rgHigh,
-			upsample:   true,
-		},
-	}
-}
-
-// copy command output to http response body using io.copy
-// (it's simpler, but may increase ttfb)
-//nolint:deadcode,unused // function may be switched later
-func cmdOutputCopy(out, cache io.Writer, pipeReader io.Reader) {
-	// set up a multiwriter to feed the command output
-	// to both cache file and http response
-	w := io.MultiWriter(out, cache)
-	// start copying!
-	if _, err := io.Copy(w, pipeReader); err != nil {
-		log.Printf("error while writing encoded output: %s\n", err)
-	}
-}
-
-// copy command output to http response manually with a buffer (should reduce ttfb)
-//nolint:deadcode,unused // function may be switched later
-func cmdOutputWrite(out, cache io.Writer, pipeReader io.ReadCloser) {
-	buffer := make([]byte, buffLen)
-	for {
-		n, err := pipeReader.Read(buffer)
-		if err != nil {
-			_ = pipeReader.Close()
-			break
-		}
-		data := buffer[0:n]
-		if _, err := out.Write(data); err != nil {
-			log.Printf("error while writing HTTP response: %s\n", err)
-		}
-		if _, err := cache.Write(data); err != nil {
-			log.Printf("error while writing cache file: %s\n", err)
-		}
-		if f, ok := out.(http.Flusher); ok {
-			f.Flush()
-		}
-		// reset buffer
-		for i := 0; i < n; i++ {
-			buffer[i] = 0
-		}
-	}
-}
-
-// pre-format the ffmpeg command with needed options
-func ffmpegCommand(filePath string, profile Profile) (*exec.Cmd, error) {
-	args := []string{
-		"-v", "0",
-		"-i", filePath,
-		"-map", "0:a:0",
-		"-vn",
-		"-b:a", fmt.Sprintf("%dk", profile.Bitrate),
-	}
-	args = append(args, profile.options...)
-
-	var aFilters []string
-	var aMetadata []string
-
-	// opus always forces output to 48kHz sampling rate, but we can still use upsampling
-	// to increase RG and alimiter's peak limiting precision, which is desirable in some
-	// cases. ffmpeg's `soxr` resampler is quite fast on x86-64: it takes around 5 seconds
-	// on my Ryzen 3600 to transcode an 8-minute FLAC with 2x upsample and RG applied.
-	// -- @spijet
-	if profile.upsample {
-		aFilters = append(aFilters, "aresample=96000:resampler=soxr")
-	}
-
-	switch profile.replayGain {
-	case rgForce:
-		aFilters = append(aFilters, ffmpegPreamp(6)...)
-		aMetadata = append(aMetadata, ffmpegStripRG()...)
-	case rgHigh:
-		// this baseline gain results in final track being +3~5dB louder
-		// than Foobar2000's default ReplayGain target volume.
-		// this makes it easier to listen to music in a car, where all other
-		// sources are usually ten thousand times louder than RG-adjusted music.
-		// -- @spijet
-		aFilters = append(aFilters, ffmpegPreamp(15)...)
-	}
-
-	if len(aFilters) > 0 {
-		args = append(args, "-af", strings.Join(aFilters, ", "))
-	}
-
-	args = append(args, aMetadata...)
-	args = append(args, "-f", profile.Format, "-")
-
-	ffmpegPath, err := exec.LookPath(ffmpeg)
-	if err != nil {
-		return nil, fmt.Errorf("find ffmpeg binary path: %w", err)
-	}
-	return exec.Command(ffmpegPath, args...), nil //nolint:gosec
-	// can't see a way for this be abused
-	// but please do let me know if you see otherwise
-}
-
-func ffmpegPreamp(dB int) []string {
-	return []string{
-		fmt.Sprintf("volume=replaygain=track:replaygain_preamp=%ddB:replaygain_noclip=0", dB),
-		"alimiter=level=disabled",
-		"asidedata=mode=delete:type=REPLAYGAIN",
-	}
-}
-
-func ffmpegStripRG() []string {
-	return []string{
-		"-metadata", "replaygain_album_gain=",
-		"-metadata", "replaygain_album_peak=",
-		"-metadata", "replaygain_track_gain=",
-		"-metadata", "replaygain_track_peak=",
-		"-metadata", "r128_album_gain=",
-		"-metadata", "r128_track_gain=",
-	}
-}
-
-func encode(out io.Writer, trackPath, cachePath string, profile Profile) error {
-	// prepare cache part file path
-	cachePartPath := fmt.Sprintf("%s.part", cachePath)
-	// prepare the command and file descriptors
-	cmd, err := ffmpegCommand(trackPath, profile)
-	if err != nil {
-		return fmt.Errorf("generate ffmpeg command: %w", err)
-	}
-	pipeReader, pipeWriter := io.Pipe()
-	cmd.Stdout = pipeWriter
-	cmd.Stderr = pipeWriter
-	// create cache part file
-	cacheFile, err := os.Create(cachePartPath)
-	if err != nil {
-		return fmt.Errorf("writing to cache file %q: %v: %w", cachePath, err, err)
-	}
-	// still unsure if buffer version (cmdOutputWrite) is any better than io.Copy-based one (cmdOutputCopy)
-	// initial goal here is to start streaming response asap, with smallest ttfb. more testing needed
-	// -- @spijet
-
-	// start up writers for cache file and http response
-	go cmdOutputWrite(out, cacheFile, pipeReader)
-	// run ffmpeg
-	if err := cmd.Run(); err != nil {
-		return fmt.Errorf("running ffmpeg: %w", err)
-	}
-	// close all pipes and flush cache part file
-	_ = pipeWriter.Close()
-	if err := cacheFile.Sync(); err != nil {
-		return fmt.Errorf("flushing %q: %w", cachePath, err)
-	}
-	_ = cacheFile.Close()
-	// rename cache part file to mark it as valid cache file
-	_ = os.Rename(cachePartPath, cachePath)
-	return nil
-}
-
-// cacheKey generates the filename for the new transcode save
-func cacheKey(sourcePath string, profileName string, profile Profile) string {
-	return fmt.Sprintf("%x-%s-%dk.%s",
-		xxhash.Sum64String(sourcePath), profileName, profile.Bitrate, profile.Format,
-	)
-}
-
-type (
-	OnInvalidProfileFunc func() error
-	OnCacheHitFunc       func(Profile, string) error
-	OnCacheMissFunc      func(Profile) (io.Writer, error)
-)
-
-type Options struct {
-	TrackPath        string
-	TrackBitrate     int
-	CachePath        string
-	ProfileName      string
-	PreferredBitrate int
-	OnInvalidProfile OnInvalidProfileFunc
-	OnCacheHit       OnCacheHitFunc
-	OnCacheMiss      OnCacheMissFunc
-}
-
-func Encode(opts Options) error {
-	profile, ok := Profiles()[opts.ProfileName]
-	if !ok {
-		return opts.OnInvalidProfile()
-	}
-	switch {
-	case opts.PreferredBitrate != 0 && opts.PreferredBitrate >= opts.TrackBitrate:
-		log.Printf("not transcoding, requested bitrate larger or equal to track bitrate\n")
-		return opts.OnInvalidProfile()
-	case opts.PreferredBitrate != 0 && opts.PreferredBitrate < opts.TrackBitrate:
-		profile.Bitrate = opts.PreferredBitrate
-		log.Printf("transcoding according to client request of %dk \n", profile.Bitrate)
-	case opts.PreferredBitrate == 0 && profile.Bitrate >= opts.TrackBitrate:
-		log.Printf("not transcoding, profile bitrate larger or equal to track bitrate\n")
-		return opts.OnInvalidProfile()
-	case opts.PreferredBitrate == 0 && profile.Bitrate < opts.TrackBitrate:
-		log.Printf("transcoding according to transcoding profile of %dk\n", profile.Bitrate)
-	}
-	cacheKey := cacheKey(opts.TrackPath, opts.ProfileName, profile)
-	cachePath := path.Join(opts.CachePath, cacheKey)
-	if fileExists(cachePath) {
-		return opts.OnCacheHit(profile, cachePath)
-	}
-	writer, err := opts.OnCacheMiss(profile)
-	if err != nil {
-		return fmt.Errorf("starting cache serve: %w", err)
-	}
-	if err := encode(writer, opts.TrackPath, cachePath, profile); err != nil {
-		return fmt.Errorf("starting transcode: %w", err)
-	}
-	return nil
-}
--- a/server/mockfs/mockfs.go
+++ b/server/mockfs/mockfs.go
@@ -148,6 +148,22 @@ func (m *MockFS) Symlink(src, dest string) {
 	}
 }

+func (m *MockFS) SetRealAudio(path string, length int, audioPath string) {
+	abspath := filepath.Join(m.dir, path)
+	if err := os.Remove(abspath); err != nil {
+		m.t.Fatalf("remove all: %v", err)
+	}
+	wd, _ := os.Getwd()
+	if err := os.Symlink(filepath.Join(wd, audioPath), abspath); err != nil {
+		m.t.Fatalf("symlink: %v", err)
+	}
+	m.SetTags(path, func(tags *Tags) error {
+		tags.RawLength = length
+		tags.RawBitrate = 0
+		return nil
+	})
+}
+
 func (m *MockFS) LogItems() {
 	m.t.Logf("\nitems")
 	var items int
@@ -337,6 +353,9 @@ type Tags struct {
 	RawAlbum       string
 	RawAlbumArtist string
 	RawGenre       string
+
+	RawBitrate int
+	RawLength  int
 }

 func (m *Tags) Title() string         { return m.RawTitle }
@@ -348,10 +367,11 @@ func (m *Tags) AlbumBrainzID() string { return "" }
 func (m *Tags) Genre() string         { return m.RawGenre }
 func (m *Tags) TrackNumber() int      { return 1 }
 func (m *Tags) DiscNumber() int       { return 1 }
-func (m *Tags) Length() int           { return 100 }
-func (m *Tags) Bitrate() int          { return 100 }
 func (m *Tags) Year() int             { return 2021 }

+func (m *Tags) Length() int  { return firstInt(100, m.RawLength) }
+func (m *Tags) Bitrate() int { return firstInt(100, m.RawBitrate) }
+
 func (m *Tags) SomeAlbum() string       { return first("Unknown Album", m.Album()) }
 func (m *Tags) SomeArtist() string      { return first("Unknown Artist", m.Artist()) }
 func (m *Tags) SomeAlbumArtist() string { return first("Unknown Artist", m.AlbumArtist(), m.Artist()) }
@@ -367,3 +387,12 @@ func first(or string, strs ...string) string {
 	}
 	return or
 }
+
+func firstInt(or int, ints ...int) int {
+	for _, int := range ints {
+		if int > 0 {
+			return int
+		}
+	}
+	return or
+}
--- a/server/server.go
+++ b/server/server.go
@@ -23,6 +23,7 @@ import (
 	"go.senan.xyz/gonic/server/scrobble"
 	"go.senan.xyz/gonic/server/scrobble/lastfm"
 	"go.senan.xyz/gonic/server/scrobble/listenbrainz"
+	"go.senan.xyz/gonic/server/transcode"
 )

 type Options struct {
@@ -84,6 +85,11 @@ func New(opts Options) (*Server, error) {

 	podcast := podcasts.New(opts.DB, opts.PodcastPath, tagger)

+	cacheTranscoder := transcode.NewCachingTranscoder(
+		transcode.NewFFmpegTranscoder(),
+		opts.CachePath,
+	)
+
 	ctrlAdmin, err := ctrladmin.New(base, sessDB, podcast)
 	if err != nil {
 		return nil, fmt.Errorf("create admin controller: %w", err)
@@ -97,6 +103,7 @@ func New(opts Options) (*Server, error) {
 		Jukebox:        &jukebox.Jukebox{},
 		Scrobblers:     []scrobble.Scrobbler{&lastfm.Scrobbler{DB: opts.DB}, &listenbrainz.Scrobbler{}},
 		Podcasts:       podcast,
+		Transcoder:     cacheTranscoder,
 	}

 	setupMisc(r, base)
@@ -222,9 +229,9 @@ func setupSubsonic(r *mux.Router, ctrl *ctrlsubsonic.Controller) {
 	r.Handle("/getSimilarSongs2{_:(?:\\.view)?}", ctrl.H(ctrl.ServeGetSimilarSongsTwo))

 	// raw
-	r.Handle("/download{_:(?:\\.view)?}", ctrl.HR(ctrl.ServeDownload))
 	r.Handle("/getCoverArt{_:(?:\\.view)?}", ctrl.HR(ctrl.ServeGetCoverArt))
 	r.Handle("/stream{_:(?:\\.view)?}", ctrl.HR(ctrl.ServeStream))
+	r.Handle("/download{_:(?:\\.view)?}", ctrl.HR(ctrl.ServeStream))

 	// browse by tag
 	r.Handle("/getAlbum{_:(?:\\.view)?}", ctrl.H(ctrl.ServeGetAlbum))
--- a/server/transcode/testdata/10s.mp3
+++ b/server/transcode/testdata/10s.mp3
--- a/server/transcode/testdata/5s.mp3
+++ b/server/transcode/testdata/5s.mp3
--- a/server/transcode/testdata/fuzz/FuzzGuessExpectedSize/1e10ee326324ee7e1f371585414e7e41d5c802b4db45ebc79abe0397fb915a0c
+++ b/server/transcode/testdata/fuzz/FuzzGuessExpectedSize/1e10ee326324ee7e1f371585414e7e41d5c802b4db45ebc79abe0397fb915a0c
@@ -0,0 +1,3 @@
+go test fuzz v1
+byte('Y')
+byte('\x05')
--- a/server/transcode/testdata/fuzz/FuzzGuessExpectedSize/3eaed9c9c8a57078033cea3337a79e1e6bfebf7e5eb7421878c911df2d78d1ac
+++ b/server/transcode/testdata/fuzz/FuzzGuessExpectedSize/3eaed9c9c8a57078033cea3337a79e1e6bfebf7e5eb7421878c911df2d78d1ac
@@ -0,0 +1,3 @@
+go test fuzz v1
+byte('\x15')
+byte('}')
--- a/server/transcode/testdata/fuzz/FuzzGuessExpectedSize/ce5db7c5c16dcb48567255c77772cacc3ca86d524cbb2d3a769248625bc3edc8
+++ b/server/transcode/testdata/fuzz/FuzzGuessExpectedSize/ce5db7c5c16dcb48567255c77772cacc3ca86d524cbb2d3a769248625bc3edc8
@@ -0,0 +1,3 @@
+go test fuzz v1
+byte('\a')
+byte('\x02')
--- a/server/transcode/transcode.go
+++ b/server/transcode/transcode.go
@@ -0,0 +1,129 @@
+// author: spijet (https://github.com/spijet/)
+// author: sentriz (https://github.com/sentriz/)
+
+//nolint:gochecknoglobals
+package transcode
+
+import (
+	"context"
+	"fmt"
+	"io"
+	"os/exec"
+	"time"
+
+	"github.com/google/shlex"
+)
+
+type Transcoder interface {
+	Transcode(ctx context.Context, profile Profile, in string) (io.ReadCloser, error)
+}
+
+var UserProfiles = map[string]Profile{
+	"mp3":      MP3,
+	"mp3_rg":   MP3RG,
+	"opus_car": OpusCar,
+	"opus":     Opus,
+	"opus_rg":  OpusRG,
+}
+
+// Store as simple strings, since we may let the user provide their own profiles soon
+var (
+	MP3   = NewProfile("audio/mpeg", 128, `ffmpeg -v 0 -i <file> -ss <seek> -map 0:a:0 -vn -b:a <bitrate> -c:a libmp3lame -af "volume=replaygain=track:replaygain_preamp=6dB:replaygain_noclip=0, alimiter=level=disabled, asidedata=mode=delete:type=REPLAYGAIN" -metadata replaygain_album_gain= -metadata replaygain_album_peak= -metadata replaygain_track_gain= -metadata replaygain_track_peak= -metadata r128_album_gain= -metadata r128_track_gain= -f mp3 -`)
+	MP3RG = NewProfile("audio/mpeg", 128, `ffmpeg -v 0 -i <file> -ss <seek> -map 0:a:0 -vn -b:a <bitrate> -c:a libmp3lame -af "volume=replaygain=track:replaygain_preamp=6dB:replaygain_noclip=0, alimiter=level=disabled, asidedata=mode=delete:type=REPLAYGAIN" -metadata replaygain_album_gain= -metadata replaygain_album_peak= -metadata replaygain_track_gain= -metadata replaygain_track_peak= -metadata r128_album_gain= -metadata r128_track_gain= -f mp3 -`)
+
+	// this sets a baseline gain which results in the final track being +3~5dB louder than
+	// Foobar2000's default ReplayGain target volume.
+	// this makes it easier to listen to music in a car, where all other
+	// sources are usually ten thousand times louder than RG-adjusted music.
+	//
+	// opus always forces output to 48kHz sampling rate, but we can still use upsampling
+	// to increase RG and alimiter's peak limiting precision, which is desirable in some
+	// cases. ffmpeg's `soxr` resampler is quite fast on x86-64: it takes around 5 seconds
+	// on my Ryzen 3600 to transcode an 8-minute FLAC with 2x upsample and RG applied.
+	//
+	// -- @spijet
+	OpusCar = NewProfile("audio/ogg", 96, `ffmpeg -v 0 -i <file> -ss <seek> -map 0:a:0 -vn -b:a <bitrate> -c:a libopus -vbr on -af "aresample=96000:resampler=soxr, volume=replaygain=track:replaygain_preamp=15dB:replaygain_noclip=0, alimiter=level=disabled, asidedata=mode=delete:type=REPLAYGAIN" -f opus -`)
+	Opus    = NewProfile("audio/ogg", 96, `ffmpeg -v 0 -i <file> -ss <seek> -map 0:a:0 -vn -b:a <bitrate> -c:a libopus -vbr on -af "volume=replaygain=track:replaygain_preamp=6dB:replaygain_noclip=0, alimiter=level=disabled, asidedata=mode=delete:type=REPLAYGAIN" -metadata replaygain_album_gain= -metadata replaygain_album_peak= -metadata replaygain_track_gain= -metadata replaygain_track_peak= -metadata r128_album_gain= -metadata r128_track_gain= -f opus -`)
+	OpusRG  = NewProfile("audio/ogg", 96, `ffmpeg -v 0 -i <file> -ss <seek> -map 0:a:0 -vn -b:a <bitrate> -c:a libopus -vbr on -af "volume=replaygain=track:replaygain_preamp=6dB:replaygain_noclip=0, alimiter=level=disabled, asidedata=mode=delete:type=REPLAYGAIN" -metadata replaygain_album_gain= -metadata replaygain_album_peak= -metadata replaygain_track_gain= -metadata replaygain_track_peak= -metadata r128_album_gain= -metadata r128_track_gain= -f opus -`)
+
+	PCM16le = NewProfile("audio/wav", 0, `ffmpeg -v 0 -i <file> -ss <seek> -c:a pcm_s16le -ac 2 -f s16le -`)
+)
+
+type BitRate int // kb/s
+
+type Profile struct {
+	bitrate BitRate // the default bitrate, but the user can request a different one
+	seek    time.Duration
+	mime    string
+	exec    string
+}
+
+func (p *Profile) BitRate() BitRate    { return p.bitrate }
+func (p *Profile) Seek() time.Duration { return p.seek }
+func (p *Profile) MIME() string        { return p.mime }
+
+func NewProfile(mime string, bitrate BitRate, exec string) Profile {
+	return Profile{mime: mime, bitrate: bitrate, exec: exec}
+}
+
+func WithBitrate(p Profile, bitRate BitRate) Profile {
+	p.bitrate = bitRate
+	return p
+}
+func WithSeek(p Profile, seek time.Duration) Profile {
+	p.seek = seek
+	return p
+}
+
+var ErrNoProfileParts = fmt.Errorf("not enough profile parts")
+
+func parseProfile(profile Profile, in string) (string, []string, error) {
+	parts, err := shlex.Split(profile.exec)
+	if err != nil {
+		return "", nil, fmt.Errorf("split command: %w", err)
+	}
+	if len(parts) == 0 {
+		return "", nil, ErrNoProfileParts
+	}
+	name, err := exec.LookPath(parts[0])
+	if err != nil {
+		return "", nil, fmt.Errorf("find name: %w", err)
+	}
+
+	var args []string
+	for _, p := range parts[1:] {
+		switch p {
+		case "<file>":
+			args = append(args, in)
+		case "<seek>":
+			args = append(args, fmt.Sprintf("%dus", profile.Seek().Microseconds()))
+		case "<bitrate>":
+			args = append(args, fmt.Sprintf("%dk", profile.BitRate()))
+		default:
+			args = append(args, p)
+		}
+	}
+
+	return name, args, nil
+}
+
+// GuessExpectedSize guesses how big the transcoded file will be in bytes.
+// Handy if we want to send a Content-Length header to the client before
+// the transcode has finished. This way, clients like DSub can render their
+// scrub bar and duration as the track is streaming.
+//
+// The estimate should overshoot a bit (2s in this case) otherwise some HTTP
+// clients will shit their trousers given some unexpected bytes.
+func GuessExpectedSize(profile Profile, length time.Duration) int {
+	if length == 0 {
+		return 0
+	}
+
+	bytesPerSec := int(profile.BitRate() * 1000 / 8)
+
+	var guess int
+	guess += bytesPerSec * int(length.Seconds()-profile.seek.Seconds())
+	guess += bytesPerSec * 2 // 2s pading
+	guess += 10000           // 10kb byte padding
+	return guess
+}
--- a/server/transcode/transcode_test.go
+++ b/server/transcode/transcode_test.go
@@ -0,0 +1,47 @@
+//go:build go1.18
+// +build go1.18
+
+package transcode_test
+
+import (
+	"context"
+	"io"
+	"testing"
+	"time"
+
+	"github.com/matryer/is"
+	"go.senan.xyz/gonic/server/transcode"
+)
+
+// FuzzGuessExpectedSize makes sure all of our profile's estimated transcode
+// file sizes are slightly bigger than the real thing.
+func FuzzGuessExpectedSize(f *testing.F) {
+	var profiles []transcode.Profile
+	for _, v := range transcode.UserProfiles {
+		profiles = append(profiles, v)
+	}
+
+	type track struct {
+		path   string
+		length time.Duration
+	}
+	var tracks []track
+	tracks = append(tracks, track{"testdata/5s.mp3", 5 * time.Second})
+	tracks = append(tracks, track{"testdata/10s.mp3", 10 * time.Second})
+
+	tr := transcode.NewFFmpegTranscoder()
+	f.Fuzz(func(t *testing.T, pseed uint8, tseed uint8) {
+		is := is.New(t)
+		profile := profiles[int(pseed)%len(profiles)]
+		track := tracks[int(tseed)%len(tracks)]
+
+		sizeGuess := transcode.GuessExpectedSize(profile, track.length)
+
+		reader, err := tr.Transcode(context.Background(), profile, track.path)
+		is.NoErr(err)
+
+		actual, err := io.ReadAll(reader)
+		is.NoErr(err)
+		is.True(sizeGuess > len(actual))
+	})
+}
--- a/server/transcode/transcoder_caching.go
+++ b/server/transcode/transcoder_caching.go
@@ -0,0 +1,65 @@
+package transcode
+
+import (
+	"context"
+	"crypto/md5"
+	"fmt"
+	"io"
+	"os"
+	"path/filepath"
+
+	"go.senan.xyz/gonic/iout"
+)
+
+const perm = 0644
+
+type CachingTranscoder struct {
+	cachePath  string
+	transcoder Transcoder
+}
+
+var _ Transcoder = (*CachingTranscoder)(nil)
+
+func NewCachingTranscoder(t Transcoder, cachePath string) *CachingTranscoder {
+	return &CachingTranscoder{transcoder: t, cachePath: cachePath}
+}
+
+func (t *CachingTranscoder) Transcode(ctx context.Context, profile Profile, in string) (io.ReadCloser, error) {
+	if err := os.MkdirAll(t.cachePath, perm^0111); err != nil {
+		return nil, fmt.Errorf("make cache path: %w", err)
+	}
+
+	name, args, err := parseProfile(profile, in)
+	if err != nil {
+		return nil, fmt.Errorf("split command: %w", err)
+	}
+
+	key := cacheKey(name, args)
+	path := filepath.Join(t.cachePath, key)
+
+	cf, err := os.OpenFile(path, os.O_RDWR|os.O_CREATE, 0644)
+	if err != nil {
+		return nil, fmt.Errorf("open cache file: %w", err)
+	}
+	if i, err := cf.Stat(); err == nil && i.Size() > 0 {
+		return cf, nil
+	}
+
+	out, err := t.transcoder.Transcode(ctx, profile, in)
+	if err != nil {
+		return nil, fmt.Errorf("internal transcode: %w", err)
+	}
+
+	return iout.NewTeeCloser(out, cf), nil
+}
+
+func cacheKey(cmd string, args []string) string {
+	// the cache is invalid whenever transcode command (which includes the
+	// absolute filepath, bit rate args, replay gain args, etc.) changes
+	sum := md5.New()
+	_, _ = io.WriteString(sum, cmd)
+	for _, arg := range args {
+		_, _ = io.WriteString(sum, arg)
+	}
+	return fmt.Sprintf("%x", sum.Sum(nil))
+}
--- a/server/transcode/transcoder_ffmpeg.go
+++ b/server/transcode/transcoder_ffmpeg.go
@@ -0,0 +1,39 @@
+package transcode
+
+import (
+	"context"
+	"fmt"
+	"io"
+	"os/exec"
+)
+
+type FFmpegTranscoder struct{}
+
+var _ Transcoder = (*FFmpegTranscoder)(nil)
+
+func NewFFmpegTranscoder() *FFmpegTranscoder {
+	return &FFmpegTranscoder{}
+}
+
+var ErrFFmpegExit = fmt.Errorf("ffmpeg exited with non 0 status code")
+
+func (*FFmpegTranscoder) Transcode(ctx context.Context, profile Profile, in string) (io.ReadCloser, error) {
+	name, args, err := parseProfile(profile, in)
+	if err != nil {
+		return nil, fmt.Errorf("split command: %w", err)
+	}
+
+	preader, pwriter := io.Pipe()
+
+	cmd := exec.CommandContext(ctx, name, args...)
+	cmd.Stdout = pwriter
+	if err := cmd.Start(); err != nil {
+		return nil, fmt.Errorf("starting cmd: %w", err)
+	}
+
+	go func() {
+		_ = pwriter.CloseWithError(cmd.Wait())
+	}()
+
+	return preader, nil
+}
--- a/server/transcode/transcoder_none.go
+++ b/server/transcode/transcoder_none.go
@@ -0,0 +1,19 @@
+package transcode
+
+import (
+	"context"
+	"io"
+	"os"
+)
+
+type NoneTranscoder struct{}
+
+var _ Transcoder = (*NoneTranscoder)(nil)
+
+func NewNoneTranscoder() *NoneTranscoder {
+	return &NoneTranscoder{}
+}
+
+func (*NoneTranscoder) Transcode(ctx context.Context, _ Profile, in string) (io.ReadCloser, error) {
+	return os.Open(in)
+}