From: Marcel van der Veldt <m.vanderveldt@outlook.com>
Date: Thu, 11 Dec 2025 23:40:33 +0000 (+0100)
Subject: Fix for some announcements issues
X-Git-Url: https://git.kitaultman.com/?a=commitdiff_plain;h=7ca9fc91d4d070d3e2a4f0845fec4a8a8b7b12fa;p=music-assistant-server.git

Fix for some announcements issues
---

diff --git a/music_assistant/controllers/players/player_controller.py b/music_assistant/controllers/players/player_controller.py
index 9d55e87b..85d36c96 100644
--- a/music_assistant/controllers/players/player_controller.py
+++ b/music_assistant/controllers/players/player_controller.py
@@ -1965,8 +1965,8 @@ class PlayerController(CoreController):
         await self.wait_for_state(
             player,
             PlaybackState.IDLE,
-            timeout=announcement.duration + 6,
-            minimal_time=float(announcement.duration),
+            timeout=announcement.duration + 10,
+            minimal_time=float(announcement.duration) + 2,
         )
         self.logger.debug(
             "Announcement to player %s - restore previous state...", player.display_name
diff --git a/music_assistant/controllers/streams/streams_controller.py b/music_assistant/controllers/streams/streams_controller.py
index b23620b8..bd3dadf4 100644
--- a/music_assistant/controllers/streams/streams_controller.py
+++ b/music_assistant/controllers/streams/streams_controller.py
@@ -1210,37 +1210,71 @@ class StreamsController(CoreController):
         pre_announce_url: str = ANNOUNCE_ALERT_FILE,
     ) -> AsyncGenerator[bytes, None]:
         """Get the special announcement stream."""
-        filter_params = ["loudnorm=I=-10:LRA=11:TP=-2"]
-
-        if pre_announce:
-            # Note: TTS URLs might take a while to load cause the actual data are often generated
-            # asynchronously by the TTS provider. If we ask ffmpeg to mix the pre-announce, it will
-            # wait until it reads the TTS data, so the whole stream will be delayed. It is much
-            # faster to first play the pre-announce using a separate ffmpeg stream, and only
-            # afterwards play the TTS itself.
-            #
-            # For this to be effective the player itself needs to be able to start playback fast.
-            # Finally, if the output_format is non-PCM, raw concatenation can be problematic.
-            # So far players seem to tolerate this, but it might break some player in the future.
+        announcement_data: asyncio.Queue[bytes | None] = asyncio.Queue(10)
+        # we are doing announcement in PCM first to avoid multiple encodings
+        # when mixing pre-announce and announcement
+        # also we have to deal with some TTS sources being super slow in delivering audio
+        # so we take an approach where we start fetching the announcement in the background
+        # while we can already start playing the pre-announce sound (if any)
+
+        pcm_format = (
+            output_format
+            if output_format.content_type.is_pcm()
+            else AudioFormat(
+                sample_rate=output_format.sample_rate,
+                content_type=ContentType.PCM_S16LE,
+                bit_depth=16,
+                channels=output_format.channels,
+            )
+        )
 
+        async def fetch_announcement() -> None:
+            fmt = announcement_url.rsplit(".")[-1]
             async for chunk in get_ffmpeg_stream(
-                audio_input=pre_announce_url,
-                input_format=AudioFormat(content_type=ContentType.try_parse(pre_announce_url)),
-                output_format=output_format,
-                filter_params=filter_params,
-                chunk_size=get_chunksize(output_format, 1),
+                audio_input=announcement_url,
+                input_format=AudioFormat(content_type=ContentType.try_parse(fmt)),
+                output_format=pcm_format,
+                chunk_size=get_chunksize(pcm_format, 1),
             ):
+                await announcement_data.put(chunk)
+            await announcement_data.put(None)  # signal end of stream
+
+        self.mass.create_task(fetch_announcement())
+
+        async def _announcement_stream() -> AsyncGenerator[bytes, None]:
+            """Generate the PCM audio stream for the announcement + optional pre-announce."""
+            if pre_announce:
+                async for chunk in get_ffmpeg_stream(
+                    audio_input=pre_announce_url,
+                    input_format=AudioFormat(content_type=ContentType.try_parse(pre_announce_url)),
+                    output_format=pcm_format,
+                    chunk_size=get_chunksize(pcm_format, 1),
+                ):
+                    yield chunk
+            # pad silence while we're waiting for the announcement to be ready
+            while announcement_data.empty():
+                yield b"\0" * int(
+                    pcm_format.sample_rate * (pcm_format.bit_depth / 8) * pcm_format.channels
+                )
+                await asyncio.sleep(0.1)
+            # stream announcement
+            while True:
+                announcement_chunk = await announcement_data.get()
+                if announcement_chunk is None:
+                    break
+                yield announcement_chunk
+
+        if output_format == pcm_format:
+            # no need to re-encode, just yield the raw PCM stream
+            async for chunk in _announcement_stream():
                 yield chunk
+            return
 
-        # work out output format/details
-        fmt = announcement_url.rsplit(".")[-1]
-        audio_format = AudioFormat(content_type=ContentType.try_parse(fmt))
+        # stream final announcement in requested output format
         async for chunk in get_ffmpeg_stream(
-            audio_input=announcement_url,
-            input_format=audio_format,
+            audio_input=_announcement_stream(),
+            input_format=pcm_format,
             output_format=output_format,
-            filter_params=filter_params,
-            chunk_size=get_chunksize(output_format, 1),
         ):
             yield chunk