introduce audio and video filtergraph customization

Author: hheimbuerger

README.md

@@ -119,6 +119,13 @@ vod:
         - "-tune:v=ull"   # can be passed either as combined args, and will be split
         - "-rc:v"         # or parameter ...
         - "cbr"           # ... and value on separate lines
+      # Optional filtergraph, start each chain with the special `[vin]` pad and end them in `[vout]`
+      filtergraph:
+        - "[vin]split=2[v1][v2]"                # duplicate input
+        - "[v1]crop=iw/2:ih:0:0,hflip[left]"     # left half mirrored horizontally
+        - "[v2]crop=iw/2:ih:iw/2:0,vflip[right]" # right half flipped vertically
+        - "[left][right]hstack[vout]"            # join halves back together
+
   # HLS-VOD segment behaviour (optional)
   segment-length: 4        # nominal segment length in seconds
   segment-offset: 1        # allowed +/- tolerance in seconds
@@ -133,6 +140,9 @@ vod:
   audio-profile:
     encoder: aac    # default "aac", but "copy" is an alternative
     bitrate: 192  # kbps
+    # Optional filtergraph, start each chain with the special `[ain]` pad and end them in `[aout]`
+    # filtergraph:
+    # - "[ain]asetrate=48000*1.5,aresample=48000[aout]" # Pitch the audio up by ~50 % (makes everyone sound like that famous mouse!)
   # If cache is enabled
   cache: true
   # If dir is empty, cache will be stored in the same directory as media source
@@ -147,6 +157,57 @@ hls-proxy:
   my_server: http://192.168.1.34:9981
 ```
 
+## Defining filter graphs on video and audio streams
+
+You can optionally define filtergraphs on video and audio profiles. This
+allows you to modify the streams during the transcoding process.
+
+If you don't specify any filtergraphs, you get the video scaled to the
+dimensions you specified and the first audio track from the input video.
+
+When you do supply a filtergraph:
+
+* start the chain at the source pads `[vin]` (video) or `[ain]` (audio)  
+* end the chain at `[vout]` or `[aout]` – these pads are what `-map` picks up
+
+Examples:
+
+```yaml
+vod:
+  video-profiles:
+    1080p:
+      width: 1920
+      height: 1080
+      bitrate: 5000
+      filtergraph:
+        - "[vin]format=pix_fmts=yuv420p[vout]"   # change pixel format to yuv420p
+```
+
+```yaml
+vod:
+  audio-profile:
+    filtergraph:
+      - "[ain][0:a:1]amix=inputs=2[aout]"   # mix second audio track into the first
+```
+
+### Implementation
+
+The transcoder always assembles a single FFmpeg `-filter_complex` that already contains **one video and one audio chain**:
+
+1. `[0:v]scale=…[vin]` – scales the first video stream and stores the result in pad `[vin]`.
+2. `[0:a]anull[ain]` – passes the first audio stream through unchanged into pad `[ain]`.
+3. If *no* extra filtergraph is supplied the code auto-adds `[vin]null[vout] ; [ain]anull[aout]` so the outputs exist.
+
+Both pads are then selected with:
+
+```sh
+-map [vout] -map [aout]?
+```
+
+`-map` tells FFmpeg exactly which streams (by pad name or by input index) should
+be written to the current output file. Being explicit prevents surprises when
+inputs carry multiple audio/video streams.
+
 ## Transcoding profiles for live streams
 
 go-transcode supports any formats that ffmpeg likes. We provide profiles out-of-the-box for h264+aac (mp4 container) for 360p, 540p, 720p and 1080p resolutions: `h264_360p`, `h264_540p`, `h264_720p` and `h264_1080p`. Profiles can have any name, but must match regex: `^[0-9A-Za-z_-]+$`

hlsvod/transcode.go

@@ -33,11 +33,13 @@ type VideoProfile struct {
 	Profile   string
 	Level     string
 	ExtraArgs []string
+	FilterGraph []string
 }
 
 type AudioProfile struct {
 	Encoder string // audio encoder (e.g., "aac", "copy", "libopus")
 	Bitrate int    // in kilobytes (0 means use encoder default)
+	FilterGraph []string // optional audio filtergraph chains
 }
 
 // returns a channel, that delivers name of the segments as they are encoded
@@ -86,17 +88,45 @@ func TranscodeSegments(ctx context.Context, ffmpegBinary string, config Transcod
 		"-sn", // No subtitles
 	}...)
 
-	// Video specs
+	// Filtergraph (scaling + optional user graph)
 	if config.VideoProfile != nil {
 		profile := config.VideoProfile
 
-		var scale string
+		// Build scale expression producing [vin] source pad
+		var scaleExpr string
 		if profile.Width >= profile.Height {
-			scale = fmt.Sprintf("scale=-2:%d", profile.Height)
+			scaleExpr = fmt.Sprintf("[0:v]scale=-2:%d[vin]", profile.Height)
 		} else {
-			scale = fmt.Sprintf("scale=%d:-2", profile.Width)
+			scaleExpr = fmt.Sprintf("[0:v]scale=%d:-2[vin]", profile.Width)
 		}
 
+		// Source audio pad
+		audioIn := "[0:a]anull[ain]"
+
+		graphParts := []string{scaleExpr, audioIn}
+
+            // Video filters
+            if len(profile.FilterGraph) > 0 {
+                graphParts = append(graphParts, profile.FilterGraph...)
+            } else {
+                graphParts = append(graphParts, "[vin]null[vout]")
+            }
+            // Audio filters
+            if config.AudioProfile != nil && len(config.AudioProfile.FilterGraph) > 0 {
+                graphParts = append(graphParts, config.AudioProfile.FilterGraph...)
+            } else {
+                graphParts = append(graphParts, "[ain]anull[aout]")
+            }
+		combinedFG := strings.Join(graphParts, ";")
+		// Add filter graph and explicit stream mapping (video & audio)
+		args = append(args, "-filter_complex", combinedFG)
+		args = append(args, "-map", "[vout]", "-map", "[aout]?")
+	}
+
+	// Video specs
+	if config.VideoProfile != nil {
+		profile := config.VideoProfile
+
 		// apply defaults if empty
 		encoder := profile.Encoder
 		if encoder == "" {
@@ -116,7 +146,6 @@ func TranscodeSegments(ctx context.Context, ffmpegBinary string, config Transcod
 		}
 
 		args = append(args, []string{
-			"-vf", scale,
 			"-c:v", encoder,
 			"-preset", preset,
 			"-profile:v", prof,

internal/api/hlsvod.go

@@ -144,18 +144,20 @@ func (a *ApiManagerCtx) HlsVod(r chi.Router) {
 				SegmentPrefix: profileID,
 
 				VideoProfile: &hlsvod.VideoProfile{
-					Width:   profile.Width,
-					Height:  profile.Height,
-					Bitrate: profile.Bitrate,
-					Encoder:   profile.Encoder,
-					Preset:  profile.Preset,
-					Profile: profile.Profile,
-					Level:   profile.Level,
-					ExtraArgs: profile.ExtraArgs,
+					Width:       profile.Width,
+					Height:      profile.Height,
+					Bitrate:     profile.Bitrate,
+					Encoder:     profile.Encoder,
+					Preset:      profile.Preset,
+					Profile:     profile.Profile,
+					Level:       profile.Level,
+					FilterGraph: profile.FilterGraph,
+					ExtraArgs:   profile.ExtraArgs,
 				},
 				VideoKeyframes: a.config.Vod.VideoKeyframes,
 				AudioProfile: &hlsvod.AudioProfile{
 					Bitrate: a.config.Vod.AudioProfile.Bitrate,
+					FilterGraph: a.config.Vod.AudioProfile.FilterGraph,
 				},
 
 				SegmentLength:    a.config.Vod.SegmentLength,

internal/config/config.go

@@ -54,11 +54,13 @@ type VideoProfile struct {
 	Profile   string   `mapstructure:"profile"`
 	Level     string   `mapstructure:"level"`
 	ExtraArgs []string `mapstructure:"extra-args"`
+	FilterGraph []string `mapstructure:"filtergraph"`
 }
 
 type AudioProfile struct {
-	Encoder string `mapstructure:"encoder"`
-	Bitrate int    `mapstructure:"bitrate"` // in kilobytes
+	Encoder string   `mapstructure:"encoder"`
+	Bitrate int      `mapstructure:"bitrate"` // in kilobytes
+	FilterGraph []string `mapstructure:"filtergraph"`
 }
 
 type VOD struct {