config reshape

pi/bin/audio-only-publisher.sh

@@ -2,26 +2,74 @@
 set -euo pipefail
 set +H
 
-ENV_FILE="${VIDEO_ENV_FILE:-/var/lib/roverd/video.env}"
+ENV_FILE="${ROVERD_MEDIA_ENV_FILE:-/var/lib/roverd/media.env}"
 
 if [[ ! -f "$ENV_FILE" ]]; then
 	echo "Environment file ${ENV_FILE} missing; cannot publish audio" >&2
 	exit 1
 fi
 
-# shellcheck disable=SC1090
-source "$ENV_FILE"
-AUDIO_ENABLE="${AUDIO_ENABLE:-0}"
-if [[ "${AUDIO_ENABLE}" -ne 1 ]]; then
+# Load KEY=VALUE pairs from media.env without evaluating shell syntax. SRT URLs
+# contain characters such as '&' and '#!', so sourcing this file would treat a
+# data file as code and can split a valid URL into shell control operators.
+load_env_file() {
+	local content=""
+
+	if [[ -r "$ENV_FILE" ]]; then
+		content="$(cat "$ENV_FILE")"
+	elif command -v sudo >/dev/null 2>&1; then
+		content="$(sudo -n cat "$ENV_FILE" 2>/dev/null || true)"
+	fi
+
+	if [[ -z "$content" ]]; then
+		echo "Cannot read ${ENV_FILE} (permission denied). Run as a user that can read it, or allow sudo -n for cat." >&2
+		exit 1
+	fi
+
+	local line key val
+	while IFS= read -r line || [[ -n "$line" ]]; do
+		[[ "$line" =~ ^[[:space:]]*$ ]] && continue
+		[[ "$line" =~ ^[[:space:]]*# ]] && continue
+
+		if [[ "$line" =~ ^[[:space:]]*export[[:space:]]+([A-Za-z_][A-Za-z0-9_]*)=(.*)$ ]]; then
+			key="${BASH_REMATCH[1]}"
+			val="${BASH_REMATCH[2]}"
+		elif [[ "$line" =~ ^[[:space:]]*([A-Za-z_][A-Za-z0-9_]*)=(.*)$ ]]; then
+			key="${BASH_REMATCH[1]}"
+			val="${BASH_REMATCH[2]}"
+		else
+			continue
+		fi
+
+		val="${val#${val%%[![:space:]]*}}"
+		val="${val%${val##*[![:space:]]}}"
+		if [[ "$val" =~ ^\".*\"$ ]]; then
+			val="${val:1:${#val}-2}"
+		elif [[ "$val" =~ ^\'.*\'$ ]]; then
+			val="${val:1:${#val}-2}"
+		fi
+
+		printf -v "$key" '%s' "$val"
+		export "$key"
+	done <<< "$content"
+}
+
+load_env_file
+: "${ROVERD_AUDIO_CAPTURE_ENABLE:?ROVERD_AUDIO_CAPTURE_ENABLE not set in ${ENV_FILE}}"
+if [[ "${ROVERD_AUDIO_CAPTURE_ENABLE}" -ne 1 ]]; then
 	echo "Audio capture disabled; skipping audio-only publisher" >&2
 	exit 0
 fi
-: "${AUDIO_PUBLISH_URL:?AUDIO_PUBLISH_URL not set in ${ENV_FILE}}"
+: "${ROVERD_AUDIO_CAPTURE_PUBLISH_URL:?ROVERD_AUDIO_CAPTURE_PUBLISH_URL not set in ${ENV_FILE}}"
+: "${ROVERD_AUDIO_CAPTURE_DEVICE:?ROVERD_AUDIO_CAPTURE_DEVICE not set in ${ENV_FILE}}"
+: "${ROVERD_AUDIO_CAPTURE_SAMPLE_RATE:?ROVERD_AUDIO_CAPTURE_SAMPLE_RATE not set in ${ENV_FILE}}"
+: "${ROVERD_AUDIO_CAPTURE_CHANNELS:?ROVERD_AUDIO_CAPTURE_CHANNELS not set in ${ENV_FILE}}"
+: "${ROVERD_AUDIO_CAPTURE_BITRATE:?ROVERD_AUDIO_CAPTURE_BITRATE not set in ${ENV_FILE}}"
 
-AUDIO_DEVICE="${AUDIO_DEVICE:-hw:0,0}"
+CAPTURE_DEVICE="${ROVERD_AUDIO_CAPTURE_DEVICE}"
 
-# The old 65,536-byte ALSA buffer represented about 171 ms before ffmpeg could
-# even publish the microphone audio:
+# At the default 48 kHz stereo S32_LE capture format, the old 65,536-byte ALSA
+# buffer represented about 171 ms before ffmpeg could publish microphone audio:
 #   65,536 / (48,000 samples * 2 channels * 4 bytes) = 0.1707 seconds
 # Keep the defaults much smaller because this publisher feeds an interactive
 # rover stream, where late-but-smooth audio is less useful than fresher audio.
@@ -51,8 +99,8 @@ run_pipeline() {
 		# format and so the published audio stays full-band stereo before the
 		# Opus encoder sees it.
 		-f s32le
-		-ar 48000
-		-ac 2
+		-ar "${ROVERD_AUDIO_CAPTURE_SAMPLE_RATE}"
+		-ac "${ROVERD_AUDIO_CAPTURE_CHANNELS}"
 		-i pipe:0
 
 		# Keep the known-required microphone boost, but remove the old
@@ -64,9 +112,9 @@ run_pipeline() {
 		# 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
+		-b:a "${ROVERD_AUDIO_CAPTURE_BITRATE}"
+		-ar:a "${ROVERD_AUDIO_CAPTURE_SAMPLE_RATE}"
+		-ac:a "${ROVERD_AUDIO_CAPTURE_CHANNELS}"
 
 		# Use the audio profile for quality. Latency is still controlled by the
 		# 20 ms Opus frame size and the low-buffering capture/publish options
@@ -82,13 +130,13 @@ run_pipeline() {
 		-muxdelay 0
 		-muxpreload 0
 		-f mpegts
-		"${AUDIO_PUBLISH_URL}"
+		"${ROVERD_AUDIO_CAPTURE_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.
+	# The capture format comes from roverd's normalized media config. Keeping
+	# arecord and ffmpeg on the same env-backed values prevents the publisher
+	# script from quietly disagreeing with roverd.yaml about sample rate or
+	# channel count.
 	#
 	# The buffer and period are byte counts because arecord interprets -B/-F in
 	# microseconds only when the value has an explicit time suffix. Keeping them
@@ -96,7 +144,7 @@ run_pipeline() {
 	# defaults are roughly 43 ms total buffer and 2.7 ms wakeup periods at
 	# 48 kHz stereo S32_LE, which removes about 128 ms of avoidable capture
 	# latency compared with the old 65,536-byte buffer.
-	arecord -D "${AUDIO_DEVICE}" -f S32_LE -c 2 -r 48000 -B "${AUDIO_ALSA_BUFFER_BYTES}" -F "${AUDIO_ALSA_PERIOD_BYTES}" -q -t raw \
+	arecord -D "${CAPTURE_DEVICE}" -f S32_LE -c "${ROVERD_AUDIO_CAPTURE_CHANNELS}" -r "${ROVERD_AUDIO_CAPTURE_SAMPLE_RATE}" -B "${AUDIO_ALSA_BUFFER_BYTES}" -F "${AUDIO_ALSA_PERIOD_BYTES}" -q -t raw \
 		| "${FFMPEG_BIN_PATH}" "${ffmpeg_args[@]}"
 }