trying stereo audio!

pi/bin/audio-only-publisher.sh

@@ -19,8 +19,6 @@ fi
 : "${AUDIO_PUBLISH_URL:?AUDIO_PUBLISH_URL not set in ${ENV_FILE}}"
 
 AUDIO_DEVICE="${AUDIO_DEVICE:-hw:0,0}"
-AUDIO_RATE="${AUDIO_RATE:-48000}"
-AUDIO_CHANNELS="${AUDIO_CHANNELS:-2}"
 
 if [[ -n "${FFMPEG_BIN:-}" ]]; then
 	FFMPEG_BIN_PATH="$FFMPEG_BIN"
@@ -32,27 +30,50 @@ else
 fi
 
 run_pipeline() {
-	arecord -D "${AUDIO_DEVICE}" -f S32_LE -c "${AUDIO_CHANNELS}" -r "${AUDIO_RATE}" -B 65536 -F 2048 -q -t raw \
+	local ffmpeg_args=(
-		| "${FFMPEG_BIN_PATH}" \
+		-hide_banner
-			-hide_banner \
+		-loglevel warning
-			-loglevel warning \
+		-fflags nobuffer
-			-fflags nobuffer \
+		-rtbufsize 0
-			-rtbufsize 0 \
+		-thread_queue_size 4096
-			-thread_queue_size 4096 \
+
-			-f s32le \
+		# Match the raw ALSA stream exactly so ffmpeg does not guess the pipe
-			-ar "${AUDIO_RATE}" \
+		# format and so the published audio stays full-band stereo before the
-			-ac "${AUDIO_CHANNELS}" \
+		# Opus encoder sees it.
-			-i pipe:0 \
+		-f s32le
-			-af "aresample=16000,pan=mono|c0=0.5*FL+0.5*FR,volume=25dB" \
+		-ar 48000
-			-c:a libopus \
+		-ac 2
-			-b:a 24000 \
+		-i pipe:0
-			-ar:a 16000 \
+
-			-ac:a 1 \
+		# Keep the known-required microphone boost, but remove the old
-			-application lowdelay \
+		# resample/downmix filter. That old filter threw away stereo and
-			-frame_duration 20 \
+		# limited the stream to narrow 16 kHz mono before encoding.
-			-compression_level 0 \
+		-af "volume=25dB"
-			-f mpegts \
+
+		# Opus supports up to 510 kbps. Using that ceiling keeps the stream at
+		# the highest practical quality browsers and MediaMTX can carry without
+		# trying to push raw PCM through the live path.
+		-c:a libopus
+		-b:a 510000
+		-ar:a 48000
+		-ac:a 2
+
+		# Use the audio profile for quality. Latency is still controlled by the
+		# 20 ms Opus frame size and the low-buffering capture/publish options
+		# around the encoder.
+		-application audio
+		-frame_duration 20
+		-compression_level 0
+		-f mpegts
 		"${AUDIO_PUBLISH_URL}"
+	)
+
+	# The Google Voice HAT microphone path is intentionally fixed instead of
+	# configurable. The previous env-driven sample rate/channel knobs made it
+	# easy for the rover config and the actual ffmpeg pipeline to drift apart,
+	# while the hardware path we install is always 48 kHz stereo capture.
+	arecord -D "${AUDIO_DEVICE}" -f S32_LE -c 2 -r 48000 -B 65536 -F 2048 -q -t raw \
+		| "${FFMPEG_BIN_PATH}" "${ffmpeg_args[@]}"
 }
 
 trap 'kill 0 2>/dev/null' EXIT INT TERM

pi/install_roverd.sh

@@ -321,8 +321,6 @@ VIDEO_BITRATE=2000000
 AUDIO_ENABLE=0
 AUDIO_DEVICE=hw:0,0
 AUDIO_PLAYBACK_DEVICE=forward
-AUDIO_RATE=48000
-AUDIO_CHANNELS=2
 ENV
 chown roverd:roverd /var/lib/roverd/video.env
 chmod 0640 /var/lib/roverd/video.env

pi/roverd/config.go

@@ -59,9 +59,6 @@ type AudioConfig struct {
 	CaptureEnabled bool   `yaml:"captureEnabled" json:"captureEnabled"`
 	CaptureDevice  string `yaml:"captureDevice" json:"captureDevice,omitempty"`
 	PlaybackDevice string `yaml:"playbackDevice" json:"playbackDevice,omitempty"`
-	SampleRate     int    `yaml:"sampleRate" json:"sampleRate,omitempty"`
-	Channels       int    `yaml:"channels" json:"channels,omitempty"`
-	Bitrate        int    `yaml:"bitrate" json:"bitrate,omitempty"`
 	TTSEnabled     bool   `yaml:"ttsEnabled" json:"ttsEnabled"`
 	DefaultEngine  string `yaml:"defaultEngine" json:"defaultEngine,omitempty"`
 	DefaultVoice   string `yaml:"defaultVoice" json:"defaultVoice,omitempty"`
@@ -199,9 +196,6 @@ func LoadConfig(path string) (*Config, error) {
 			CaptureEnabled: false,
 			CaptureDevice:  "rovermic",
 			PlaybackDevice: "forward",
-			SampleRate:     48000,
-			Channels:       2,
-			Bitrate:        24000,
 			TTSEnabled:     false,
 			DefaultEngine:  "flite",
 			DefaultVoice:   "rms",
@@ -364,15 +358,6 @@ func validateAudioConfig(cfg *AudioConfig) {
 	if cfg.PlaybackDevice == "" || cfg.PlaybackDevice == "default" {
 		cfg.PlaybackDevice = "forward"
 	}
-	if cfg.SampleRate <= 0 {
-		cfg.SampleRate = 48000
-	}
-	if cfg.Channels <= 0 {
-		cfg.Channels = 2
-	}
-	if cfg.Bitrate <= 0 {
-		cfg.Bitrate = 64000
-	}
 	if cfg.DefaultEngine == "" {
 		cfg.DefaultEngine = "flite"
 	}

pi/roverd/media_env.go

@@ -45,12 +45,6 @@ func UpdatePublisherEnv(media MediaConfig, audio AudioConfig) error {
 	if audioDevice == "" || audioDevice == "rovermic" {
 		audioDevice = "hw:0,0"
 	}
-	if audio.SampleRate <= 0 {
-		audio.SampleRate = 48000
-	}
-	if audio.Channels <= 0 {
-		audio.Channels = 2
-	}
 	fmt.Fprintf(&buf, "AUDIO_ENABLE=%d\n", boolToInt(audio.CaptureEnabled))
 	fmt.Fprintf(&buf, "AUDIO_DEVICE=%s\n", audioDevice)
 	playbackDevice := audio.PlaybackDevice
@@ -58,8 +52,6 @@ func UpdatePublisherEnv(media MediaConfig, audio AudioConfig) error {
 		playbackDevice = "forward"
 	}
 	fmt.Fprintf(&buf, "AUDIO_PLAYBACK_DEVICE=%s\n", playbackDevice)
-	fmt.Fprintf(&buf, "AUDIO_RATE=%d\n", audio.SampleRate)
-	fmt.Fprintf(&buf, "AUDIO_CHANNELS=%d\n", audio.Channels)
 	if err := os.WriteFile(publisherEnvPath, buf.Bytes(), 0o640); err != nil {
 		return err
 	}

pi/roverd/roverd.sample.yaml

@@ -44,9 +44,6 @@ audio:
   captureEnabled: false
   captureDevice: hw:0,0
   playbackDevice: forward
-  sampleRate: 48000
-  channels: 2
-  bitrate: 24000
   ttsEnabled: false
   defaultEngine: flite
   defaultVoice: rms