optimize for latency

README.md

@@ -138,13 +138,28 @@ sudo systemctl status auracast-frontend
 
 If you want to run the services as a specific user, edit the `User=` line in the service files accordingly.
 
-# install port audio so it can see pipewire devices on raspian
+# Setup the audio system
 sudo apt remove -y libportaudio2 portaudio19-dev libportaudiocpp0
+echo "y" | rpi-update stable
+
+
 
 sudo apt update
-sudo apt install --no-install-recommends \
+# TODO: needed  ?
+sudo apt install pipewire wireplumber pipewire-audio-client-libraries rtkit
+mkdir -p ~/.config/pipewire/pipewire.conf.d
+cp src/service/pipewire/99-lowlatency.conf ~/.config/pipewire/pipewire.conf.d/
+
+/etc/modprobe.d/usb-audio-lowlatency.conf
+snd_usb_audio nrpacks=1
+
+sudo apt install -y cpufrequtils
+sudo cpufreq-set -g performance
+
+sudo apt install -y --no-install-recommends \
         git build-essential cmake pkg-config \
-        libasound2-dev libpulse-dev libjack-jackd2-dev
+        libasound2-dev libpulse-dev libjack-jackd2-dev jackd \
+        pipewire ethtool linuxptp
 
 git clone https://github.com/PortAudio/portaudio.git
 cd portaudio 
@@ -152,9 +167,9 @@ git checkout 9abe5fe7db729280080a0bbc1397a528cd3ce658
 rm -rf build
 cmake -S . -B build -G"Unix Makefiles" \
       -DBUILD_SHARED_LIBS=ON \
-      -DPA_USE_ALSA=ON \
-      -DPA_USE_PIPEWIRE=ON \
-      -DPA_USE_JACK=OFF
+      -DPA_USE_ALSA=OFF \
+      -DPA_USE_PULSEAUDIO=ON \
+      -DPA_USE_JACK=ON
 cmake --build build -j$(nproc)
 sudo cmake --install build          # installs to /usr/local/lib
 sudo ldconfig                        # refresh linker cache

src/auracast/multicast.py

@@ -106,6 +106,88 @@ audio_io.WaveAudioInput = ModWaveAudioInput
 # -----------------------------------------------------------------------------
 logger = logging.getLogger(__name__)
 
+# -----------------------------------------------------------------------------
+# Low-latency SoundDevice input shim
+# -----------------------------------------------------------------------------
+class ModSoundDeviceAudioInput(audio_io.ThreadedAudioInput):
+    """Audio input using sounddevice with explicit low-latency settings.
+
+    This mirrors bumble.audio.io.SoundDeviceAudioInput but requests a small
+    PortAudio blocksize and low latency to reduce buffering on PipeWire/ALSA.
+    Tunables via env:
+      - AURACAST_SD_BLOCKSIZE (default 64)
+      - AURACAST_SD_LATENCY (default 'low', can be 'default', 'high' or seconds)
+    """
+
+    def __init__(self, device_name: str, pcm_format: audio_io.PcmFormat) -> None:
+        super().__init__()
+        self._device = int(device_name) if device_name else None
+        self._pcm_format = pcm_format
+        self._stream = None
+
+    def _open(self) -> audio_io.PcmFormat:
+        import os
+        import sounddevice  # type: ignore
+
+        # Read tunables
+        try:
+            blocksize = int(os.environ.get('AURACAST_SD_BLOCKSIZE', '64'))
+        except Exception:
+            blocksize = 64
+        raw_latency = os.environ.get('AURACAST_SD_LATENCY', 'low')
+        if raw_latency in ('low', 'high', 'default'):
+            latency: typing.Union[str, float] = raw_latency
+        else:
+            try:
+                latency = float(raw_latency)
+            except Exception:
+                latency = 'low'
+
+        # Create the raw input stream with tighter buffering
+        self._stream = sounddevice.RawInputStream(
+            samplerate=self._pcm_format.sample_rate,
+            device=self._device,
+            channels=self._pcm_format.channels,
+            dtype='int16',
+            blocksize=blocksize,
+            latency=latency,
+        )
+        self._stream.start()
+
+        # Report stereo output format to match Bumble's original behavior
+        return audio_io.PcmFormat(
+            audio_io.PcmFormat.Endianness.LITTLE,
+            audio_io.PcmFormat.SampleType.INT16,
+            self._pcm_format.sample_rate,
+            2,
+        )
+
+    def _read(self, frame_size: int) -> bytes:
+        if not self._stream:
+            return b''
+        pcm_buffer, overflowed = self._stream.read(frame_size)
+        if overflowed:
+            logging.warning("input overflow")
+
+        # Duplicate mono to stereo for downstream expectations
+        if self._pcm_format.channels == 1:
+            stereo_buffer = bytearray()
+            for i in range(frame_size):
+                sample = pcm_buffer[i * 2 : i * 2 + 2]
+                stereo_buffer += sample + sample
+            return bytes(stereo_buffer)
+
+        return bytes(pcm_buffer)
+
+    def _close(self):
+        if self._stream:
+            self._stream.stop()
+            self._stream = None
+
+
+# Replace Bumble's default with the low-latency variant
+audio_io.SoundDeviceAudioInput = ModSoundDeviceAudioInput
+
 @contextlib.asynccontextmanager
 async def create_device(config: auracast_config.AuracastGlobalConfig) -> AsyncGenerator[bumble.device.Device, Any]:
     async with await bumble.transport.open_transport(config.transport) as (

src/auracast/multicast_script.py

@@ -0,0 +1,83 @@
+import logging
+import os
+import asyncio
+import aioconsole
+from auracast import multicast
+from auracast import auracast_config
+from auracast.utils.sounddevice_utils import list_usb_pw_inputs, list_network_pw_inputs
+
+
+
+if __name__ == "__main__":
+    
+    logging.basicConfig( #export LOG_LEVEL=DEBUG
+        level=os.environ.get('LOG_LEVEL', logging.INFO),
+        format='%(module)s.py:%(lineno)d %(levelname)s: %(message)s'
+    )
+    os.chdir(os.path.dirname(__file__))
+    
+    # Hint PipeWire to use a very small processing quantum for this client
+    # (we will refine after SRATE is known). Defaults target ~0.33–0.67 ms.
+    os.environ.setdefault("PIPEWIRE_LATENCY", "16/48000")
+    # Lower PulseAudio-side buffering (PortAudio backend typically goes through PA)
+    os.environ.setdefault("PULSE_LATENCY_MSEC", "1")
+    # Request smaller PortAudio blocks via sounddevice shim
+    os.environ.setdefault("AURACAST_SD_BLOCKSIZE", "32")
+    # Accepts 'low'/'high'/'default' or seconds (float). Our shim parses number strings to float.
+    os.environ.setdefault("AURACAST_SD_LATENCY", "0.0015")
+    print("USB pw inputs:")
+    usb_inputs = list_usb_pw_inputs()
+    aes67_inputs = list_network_pw_inputs()
+    for i, d in usb_inputs:
+        logging.info(f"{i}: {d['name']}  in={d['max_input_channels']}")
+    for i, d in aes67_inputs:
+        logging.info(f"{i}: {d['name']}  in={d['max_input_channels']}")
+    
+    input_sel = usb_inputs[0][0]
+
+    TRANSPORT1 = 'serial:/dev/ttyAMA3,1000000,rtscts' # transport for raspberry pi gpio header
+    TRANSPORT2 = 'serial:/dev/ttyAMA4,1000000,rtscts' # transport for raspberry pi gpio header
+    # Capture at 48 kHz to avoid PipeWire resampler latency; encode LC3 at 24 kHz
+    CAPTURE_SRATE = 48000
+    LC3_SRATE = 24000
+    OCTETS_PER_FRAME=60
+
+    # After capture rate is known, constrain client buffers precisely relative to
+    # the active capture rate (so the client quantum matches the device rate).
+    try:
+        os.environ["PIPEWIRE_LATENCY"] = f"16/{CAPTURE_SRATE}"
+    except Exception:
+        pass
+    os.environ["PULSE_LATENCY_MSEC"] = os.environ.get("PULSE_LATENCY_MSEC", "1")
+    os.environ["AURACAST_SD_BLOCKSIZE"] = os.environ.get("AURACAST_SD_BLOCKSIZE", "32")
+    os.environ["AURACAST_SD_LATENCY"] = os.environ.get("AURACAST_SD_LATENCY", "0.0015")
+
+    config = auracast_config.AuracastConfigGroup(
+        bigs = [
+            auracast_config.AuracastBigConfig(
+                iso_que_len=1,
+                audio_source=f'device:{input_sel}',
+                input_format=f"int16le,{CAPTURE_SRATE},1",
+                sampling_frequency=LC3_SRATE,
+                octets_per_frame=OCTETS_PER_FRAME,
+            ),
+            #auracast_config.AuracastBigConfigEng(),
+
+        ]
+    )
+
+    config.qos_config=auracast_config.AuracastQosHigh()
+    config.transport=TRANSPORT1
+
+    # TODO: encrypted streams are not working
+
+    config.auracast_sampling_rate_hz = LC3_SRATE
+    config.octets_per_frame = OCTETS_PER_FRAME # 32kbps@16kHz
+    #config.debug = True
+
+    multicast.run_async(
+        multicast.broadcast(
+            config,
+            config.bigs
+        )
+    )

src/auracast/server/multicast_frontend.py

@@ -309,7 +309,7 @@ else:
                         )
                     ),
                     input_format=(f"int16le,{q['rate']},1" if audio_mode == "USB/Network" else "auto"),
-                    iso_que_len=1, # TODO: this should be way less to decrease delay
+                    iso_que_len=1,
                     sampling_frequency=q['rate'],
                     octets_per_frame=q['octets'],
                 ),

src/auracast/utils/sounddevice_utils.py

@@ -0,0 +1,112 @@
+import sounddevice as sd
+import os, re, json, subprocess
+
+def devices_by_backend(backend_name: str):
+    hostapis = sd.query_hostapis()          # list of host APIs
+    # find the host API index by (case-insensitive) name match
+    try:
+        hostapi_idx = next(
+            i for i, ha in enumerate(hostapis)
+            if backend_name.lower() in ha['name'].lower()
+        )
+    except StopIteration:
+        raise ValueError(f"No host API matching {backend_name!r}. "
+                         f"Available: {[ha['name'] for ha in hostapis]}")
+    # return (global_index, device_dict) pairs filtered by that host API
+    return [(i, d) for i, d in enumerate(sd.query_devices())
+            if d['hostapi'] == hostapi_idx]
+
+def _pa_like_hostapi_index():
+    for i, ha in enumerate(sd.query_hostapis()):
+        if any(k in ha["name"] for k in ("PipeWire", "PulseAudio")):
+            return i
+    raise RuntimeError("PipeWire/PulseAudio host API not present in PortAudio.")
+
+def _pw_dump():
+    return json.loads(subprocess.check_output(["pw-dump"]))
+
+def _sd_matches_from_names(pa_idx, names):
+    names_l = {n.lower() for n in names if n}
+    out = []
+    for i, d in enumerate(sd.query_devices()):
+        if d["hostapi"] != pa_idx or d["max_input_channels"] <= 0:
+            continue
+        dn = d["name"].lower()
+        if any(n in dn for n in names_l):
+            out.append((i, d))
+    return out
+
+def list_usb_pw_inputs():
+    """
+    Return [(device_index, device_dict), ...] for PipeWire **input** nodes
+    backed by **USB** devices (excludes monitor sources).
+    """
+    pa_idx = _pa_like_hostapi_index()
+    pw = _pw_dump()
+
+    # Map device.id -> device.bus ("usb"/"pci"/"platform"/"network"/...)
+    device_bus = {}
+    for obj in pw:
+        if obj.get("type") == "PipeWire:Interface:Device":
+            props = (obj.get("info") or {}).get("props") or {}
+            device_bus[obj["id"]] = (props.get("device.bus") or "").lower()
+
+    # Collect names/descriptions of USB input nodes
+    usb_input_names = set()
+    for obj in pw:
+        if obj.get("type") != "PipeWire:Interface:Node":
+            continue
+        props = (obj.get("info") or {}).get("props") or {}
+        media = (props.get("media.class") or "").lower()
+        if "source" not in media and "stream/input" not in media:
+            continue
+        # skip monitor sources ("Monitor of ..." or *.monitor)
+        nname = (props.get("node.name") or "").lower()
+        ndesc = (props.get("node.description") or "").lower()
+        if ".monitor" in nname or "monitor" in ndesc:
+            continue
+        bus = (props.get("device.bus") or device_bus.get(props.get("device.id")) or "").lower()
+        if bus == "usb":
+            usb_input_names.add(props.get("node.description") or props.get("node.name"))
+
+    # Map to sounddevice devices on PipeWire host API
+    return _sd_matches_from_names(pa_idx, usb_input_names)
+
+def list_network_pw_inputs():
+    """
+    Return [(device_index, device_dict), ...] for PipeWire **input** nodes that
+    look like network/AES67/RTP sources (excludes monitor sources).
+    """
+    pa_idx = _pa_like_hostapi_index()
+    pw = _pw_dump()
+
+    network_input_names = set()
+    for obj in pw:
+        if obj.get("type") != "PipeWire:Interface:Node":
+            continue
+        props = (obj.get("info") or {}).get("props") or {}
+        media = (props.get("media.class") or "").lower()
+        if "source" not in media and "stream/input" not in media:
+            continue
+        nname = (props.get("node.name") or "")
+        ndesc = (props.get("node.description") or "")
+        # skip monitor sources
+        if ".monitor" in nname.lower() or "monitor" in ndesc.lower():
+            continue
+
+        # Heuristics for network/AES67/RTP
+        text = (nname + " " + ndesc).lower()
+        is_network = (
+            (props.get("device.bus") or "").lower() == "network" or
+            any(k in text for k in ("rtp", "sap", "aes67", "network", "raop", "airplay"))
+        )
+        if is_network:
+            network_input_names.add(ndesc or nname)
+
+    return _sd_matches_from_names(pa_idx, network_input_names)
+
+# Example usage:
+# for i, d in list_usb_pw_inputs():
+#     print(f"USB IN  {i}: {d['name']}  in={d['max_input_channels']}")
+# for i, d in list_network_pw_inputs():
+#     print(f"NET IN  {i}: {d['name']}  in={d['max_input_channels']}")

src/scripts/list_pw_nodes.py

@@ -1,7 +1,28 @@
 import sounddevice as sd, pprint
+from auracast.utils.sounddevice_utils import devices_by_backend, list_usb_pw_inputs, list_network_pw_inputs
+
+
 print("PortAudio library:", sd._libname)
 print("PortAudio version:", sd.get_portaudio_version())
 print("\nHost APIs:")
 pprint.pprint(sd.query_hostapis())
 print("\nDevices:")
-pprint.pprint(sd.query_devices())
+pprint.pprint(sd.query_devices())
+
+# Example: only PulseAudio devices on Linux
+print("\nOnly PulseAudio devices:")
+for i, d in devices_by_backend("PulseAudio"):
+    print(f"{i}: {d['name']}  in={d['max_input_channels']} out={d['max_output_channels']}")
+
+# Example: only PulseAudio devices on Linux
+# print("\nOnly JACK devices:")
+# for i, d in devices_by_backend("JACK"):
+#     print(f"{i}: {d['name']}  in={d['max_input_channels']} out={d['max_output_channels']}")
+
+print("Network pw inputs:")
+for i, d in list_network_pw_inputs():
+    print(f"{i}: {d['name']}  in={d['max_input_channels']}")
+
+print("USB pw inputs:")
+for i, d in list_usb_pw_inputs():
+    print(f"{i}: {d['name']}  in={d['max_input_channels']}")

src/service/pipewire/99-lowlatency.conf

@@ -0,0 +1,12 @@
+context.properties = {
+    default.clock.rate          = 48000
+    default.clock.allowed-rates = [ 48000 ]
+    default.clock.quantum       = 64         # 64/48000 ≈ 1.33 ms
+    default.clock.min-quantum   = 32
+    default.clock.max-quantum   = 128
+}
+
+stream.properties = {
+    node.latency     = "64/48000"
+    resample.quality = 0
+}