slm_open_api_package 0.2.10__tar.gz → 0.2.12__tar.gz

{slm_open_api_package-0.2.10 → slm_open_api_package-0.2.12}/PKG-INFO

@@ -1,11 +1,12 @@
 Metadata-Version: 2.3
 Name: slm_open_api_package
-Version: 0.2.10
+Version: 0.2.12
 Summary: Add your description here
 Author: Sofus Vingaard
 Author-email: Sofus Vingaard <sofus.krebsvingaard@hbkworld.com>
 Requires-Dist: requests
 Requires-Dist: numpy
+Requires-Dist: pandas
 Requires-Dist: websockets
 Requires-Dist: kaitaistruct
 Requires-Dist: packaging
@@ -13,6 +14,7 @@ Requires-Dist: pyqtgraph
 Requires-Dist: pyqt5
 Requires-Dist: pyqt5-qt5<5.15.11
 Requires-Dist: miniaudio
+Requires-Dist: soundfile
 Requires-Dist: matplotlib>=3.11.1
 Requires-Dist: cryptography>=50.0.1
 Requires-Python: >=3.12
@@ -78,7 +80,7 @@ except KeyboardInterrupt:
 	streamer.stopStream()
 ```
 
-`stopStream()` stops the measurement, removes the device stream, and writes buffered saved data. Supported saving formats are `"csv"`, `"json"`, and `"pickle"`; their extensions are added automatically.
+JSON and CSV publish complete snapshots in a background worker, at most once per second while new data arrives, rather than rewriting the file for every sample. They retain records in memory and preserve their existing layouts; full-history snapshot costs still grow with recording length. Construct `JsonDataHandler(path, snapshot_interval=2.0)` or `CsvDataHandler(path, snapshot_interval=2.0)` and attach it with `setDataHandler()` to use a different interval. `close()` waits for the worker and saves all remaining records; background write failures are reported on subsequent updates or during the final flush. WAV still appends each audio chunk without buffering the recording in memory. `stopStream()` stops the measurement, closes saving handlers, and removes the device stream. Measurement saving formats are `"csv"` and `"json"`; audio uses `"wav"`. File extensions are added automatically.
 
 ## Multiple Sequences
 
@@ -147,11 +149,22 @@ streamer = WebXiStreamHandler(
 	host="192.168.1.183",
 	sequenceID=SequenceIdEnums.MP3Signal.value,
 	mode="mp3",
+	saving="wav",
+	saving_path="saved_data/audio",
 )
+
+try:
+	streamer.startStream()
+except KeyboardInterrupt:
+	streamer.stopStream()
 ```
 
 The matching `flac` form uses `SequenceIdEnums.FLACSignal.value` and `mode="flac"`. Audio setup is enabled by the handler. The SLM can take a few seconds to initialize audio recording when its quality setting changes.
 
+WAV saving appends decoded mono samples as recording happens, updating the file header after each chunk so the file is readable before shutdown. Saving happens after decoding and calibration to pascals. Samples are divided by the fixed full-scale pressure `calibrationFactor * (2**23 - 1)` and saved as `PCM_24`, not peak-normalized per chunk. The filename includes `_fullscale-<value>Pa.wav`. To recover pressure values, multiply floating-point samples read from the WAV by that full-scale value (within 24-bit quantization accuracy). Samples outside the calibrated range produce a clipping warning; this scaling cannot repair upstream clipping. MP3 uses the decoded sample rate; the current FLAC decoder uses 65536 Hz. JSON and CSV saving are rejected for audio modes, and WAV saving is rejected for measurement modes. Run `uv sync` to install the `soundfile` dependency used for progressive WAV writing.
+
+To attach the handler directly, use `WavDataHandler("saved_data/audio", sample_rate=65536, full_scale_pa=calibrationFactor * (2**23 - 1))` from `slm_api.helpers.data_handler`, using the device's actual calibration factor. You can instead supply `sample_rate` and `full_scale_pa` to `handle()`; the stream supplies both automatically. Call `close()` when finished. Use a separate WAV handler and output path for each device. Existing float32 pressure WAVs are unchanged; this format applies to new recordings.
+
 ## Synchronized Multi-Device Stream
 
 Based on [multi_device_flac.py](src/slm_api/multi_device_flac.py), pass `multi_device=True` and start/stop the group through the synchronized helpers:
@@ -168,7 +181,7 @@ streams = [
 		mode="mp3",
 		streamName=f"Audio_{host}",
 		multi_device=True,
-		saving="csv",
+		saving="wav",
 		saving_path=f"saved_data/{host}",
 	)
 	for host in hosts
@@ -203,4 +216,4 @@ Legacy flags such as `mp3=True` remain supported, but `mode=` is preferred. Only
 
 - **No values arrive:** Confirm the requested sequences are enabled on the SLM.
 - **Audio returns no samples after first enable:** Allow the SLM a few seconds to initialize audio recording before retrying.
-- **Saved file is missing:** Call `stopStream()` or `stop_all()`; buffered files are written during shutdown.
+- **Saved file is missing:** Files are written when the handler first receives data (nonempty samples for WAV). Check the saving path and that the stream is delivering data. Still call `stopStream()` or `stop_all()` to clean up recording resources.

{slm_open_api_package-0.2.10 → slm_open_api_package-0.2.12}/README.md

@@ -58,7 +58,7 @@ except KeyboardInterrupt:
 	streamer.stopStream()
 ```
 
-`stopStream()` stops the measurement, removes the device stream, and writes buffered saved data. Supported saving formats are `"csv"`, `"json"`, and `"pickle"`; their extensions are added automatically.
+JSON and CSV publish complete snapshots in a background worker, at most once per second while new data arrives, rather than rewriting the file for every sample. They retain records in memory and preserve their existing layouts; full-history snapshot costs still grow with recording length. Construct `JsonDataHandler(path, snapshot_interval=2.0)` or `CsvDataHandler(path, snapshot_interval=2.0)` and attach it with `setDataHandler()` to use a different interval. `close()` waits for the worker and saves all remaining records; background write failures are reported on subsequent updates or during the final flush. WAV still appends each audio chunk without buffering the recording in memory. `stopStream()` stops the measurement, closes saving handlers, and removes the device stream. Measurement saving formats are `"csv"` and `"json"`; audio uses `"wav"`. File extensions are added automatically.
 
 ## Multiple Sequences
 
@@ -127,11 +127,22 @@ streamer = WebXiStreamHandler(
 	host="192.168.1.183",
 	sequenceID=SequenceIdEnums.MP3Signal.value,
 	mode="mp3",
+	saving="wav",
+	saving_path="saved_data/audio",
 )
+
+try:
+	streamer.startStream()
+except KeyboardInterrupt:
+	streamer.stopStream()
 ```
 
 The matching `flac` form uses `SequenceIdEnums.FLACSignal.value` and `mode="flac"`. Audio setup is enabled by the handler. The SLM can take a few seconds to initialize audio recording when its quality setting changes.
 
+WAV saving appends decoded mono samples as recording happens, updating the file header after each chunk so the file is readable before shutdown. Saving happens after decoding and calibration to pascals. Samples are divided by the fixed full-scale pressure `calibrationFactor * (2**23 - 1)` and saved as `PCM_24`, not peak-normalized per chunk. The filename includes `_fullscale-<value>Pa.wav`. To recover pressure values, multiply floating-point samples read from the WAV by that full-scale value (within 24-bit quantization accuracy). Samples outside the calibrated range produce a clipping warning; this scaling cannot repair upstream clipping. MP3 uses the decoded sample rate; the current FLAC decoder uses 65536 Hz. JSON and CSV saving are rejected for audio modes, and WAV saving is rejected for measurement modes. Run `uv sync` to install the `soundfile` dependency used for progressive WAV writing.
+
+To attach the handler directly, use `WavDataHandler("saved_data/audio", sample_rate=65536, full_scale_pa=calibrationFactor * (2**23 - 1))` from `slm_api.helpers.data_handler`, using the device's actual calibration factor. You can instead supply `sample_rate` and `full_scale_pa` to `handle()`; the stream supplies both automatically. Call `close()` when finished. Use a separate WAV handler and output path for each device. Existing float32 pressure WAVs are unchanged; this format applies to new recordings.
+
 ## Synchronized Multi-Device Stream
 
 Based on [multi_device_flac.py](src/slm_api/multi_device_flac.py), pass `multi_device=True` and start/stop the group through the synchronized helpers:
@@ -148,7 +159,7 @@ streams = [
 		mode="mp3",
 		streamName=f"Audio_{host}",
 		multi_device=True,
-		saving="csv",
+		saving="wav",
 		saving_path=f"saved_data/{host}",
 	)
 	for host in hosts
@@ -183,4 +194,4 @@ Legacy flags such as `mp3=True` remain supported, but `mode=` is preferred. Only
 
 - **No values arrive:** Confirm the requested sequences are enabled on the SLM.
 - **Audio returns no samples after first enable:** Allow the SLM a few seconds to initialize audio recording before retrying.
-- **Saved file is missing:** Call `stopStream()` or `stop_all()`; buffered files are written during shutdown.
+- **Saved file is missing:** Files are written when the handler first receives data (nonempty samples for WAV). Check the saving path and that the stream is delivering data. Still call `stopStream()` or `stop_all()` to clean up recording resources.

{slm_open_api_package-0.2.10 → slm_open_api_package-0.2.12}/pyproject.toml

@@ -1,12 +1,13 @@
 [project]
 name = "slm_open_api_package"
-version = "0.2.10"
+version = "0.2.12"
 description = "Add your description here"
 readme = "README.md"
 requires-python = ">=3.12"
 dependencies = [
     "requests",
     "numpy",
+    "pandas",
     "websockets",
     "kaitaistruct",
     "packaging",
@@ -14,6 +15,7 @@ dependencies = [
     "PyQt5",
     "PyQt5-Qt5<5.15.11",
     "miniaudio",
+    "soundfile",
     "matplotlib>=3.11.1",
     "cryptography>=50.0.1",
 ]

{slm_open_api_package-0.2.10 → slm_open_api_package-0.2.12}/pyproject.toml.orig

@@ -1,6 +1,6 @@
 [project]
 name = "slm_open_api_package"
-version = "0.2.10"
+version = "0.2.12"
 description = "Add your description here"
 readme = "README.md"
 authors = [
@@ -10,6 +10,7 @@ requires-python = ">=3.12"
 dependencies = [
     "requests",
     "numpy",
+    "pandas",
     "websockets",
     "kaitaistruct",
     "packaging",
@@ -17,6 +18,7 @@ dependencies = [
     "PyQt5",
     "PyQt5-Qt5<5.15.11",
     "miniaudio",
+    "soundfile",
     "matplotlib>=3.11.1",
     "cryptography>=50.0.1",
 ]

{slm_open_api_package-0.2.10 → slm_open_api_package-0.2.12}/src/slm_api/CPB_example.py

@@ -6,7 +6,8 @@ from matplotlib.animation import FuncAnimation
 from slm_api.helpers.stream_handlers import WebXiStreamHandler
 from slm_api.helpers import webxi_helper_functions as webxi_helper 
 from slm_api.enums.sequence_id_enum import SequenceIdEnums
-from slm_api.helpers.data_handler import CsvDataHandler,PickleDataHandler,JsonDataHandler,MultiDataHandler, DataHandler
+from slm_api.helpers.data_handler import CsvDataHandler,JsonDataHandler,MultiDataHandler, DataHandler
+import datetime
 
 
 host = webxi_helper.set_host(__file__)
@@ -17,8 +18,8 @@ cpb_sequence_names = ["CPBLAeq"]
 sequenceID= SequenceIdEnums.CPBLAeq.value   
 
 class PrintHandler(DataHandler):
-    def handle(self, value, **data ):
-        print(f"{value}")
+    def handle(self, timestamp, value, **data ):
+        print(f"time = {timestamp} value: {value}")
 
 
 class FigureHandler:
@@ -59,12 +60,16 @@ if __name__ == "__main__":
     # turns on the CPB eq we want to use
     webxi_helper.turn_on_cpb_leq(host, cpb_sequence_names)
 
+    filename = f'name-of-file_{datetime.datetime.now().strftime("%H%M_%m%d%Y")}'
+
+
     streamer = WebXiStreamHandler(host,
                                   sequenceID=sequenceID, 
                                   mode="cpb",
                                   sequenceNames=cpb_sequence_names, 
-                                  saving_path=r"./saved_data/test_file",
-                                  saving="json")
+                                  saving_path=f"./saved_data/{filename}",
+                                  saving="json",
+                                  time=True)
     # streamer.setDataHandler(JsonDataHandler(r"C:\Users\SKREBSVINGAA\csv_test\csv_test.txt"))
     streamer.setDataHandler(PrintHandler())
     

{slm_open_api_package-0.2.10 → slm_open_api_package-0.2.12}/src/slm_api/example_multi.py

@@ -8,6 +8,8 @@ from matplotlib.animation import FuncAnimation
 from slm_api.helpers import webxi_helper_functions as webxi_helper
 from slm_api.helpers import stream_handlers as sh
 from slm_api.helpers.data_handler import DataHandler, MultiDataHandler
+import datetime
+
 
 # UseHttps(use_https=False)
 
@@ -15,7 +17,7 @@ host = webxi_helper.set_host(__file__)
 
 # This example will stream LAeq. Add more names to the list to plot several sequences at once.
 # Incase of error make sure the sequences are enabled on the SLM.
-sequenceNames = ["LAeq", "LCeq", "LAF"]
+sequenceNames = ["LAeq", "LCeq"]
 
 class PrintHandler(DataHandler):
     def handle(self, *, name, value, **data):
@@ -70,7 +72,9 @@ if __name__ == "__main__":
     # sets the sequences to true. You can add or remove sequences at the top of the file.
     webxi_helper.turn_on_bb_leq(host, sequenceNames)
 
-    streamer = sh.WebXiStreamHandler(host=host, sequenceNames=sequenceNames, mode="multi", saving="json", saving_path=f"./saved_data/test")
+    filename = f'name-of-file_{datetime.datetime.now().strftime("%H%M_%m%d%Y")}'
+
+    streamer = sh.WebXiStreamHandler(host=host, sequenceNames=sequenceNames, time=True, mode="multi", saving="csv", saving_path=f"./saved_data/{filename}")
     fig = FigHandler(sequenceNames)
     streamer.setDataHandler(MultiDataHandler(PrintHandler(), fig))
     fig.startAnimation()

{slm_open_api_package-0.2.10 → slm_open_api_package-0.2.12}/src/slm_api/fast_log_example.py

@@ -36,6 +36,7 @@ from slm_api.helpers import webxi_helper_functions as webxi_helper
 # enums
 from slm_api.enums.sequence_id_enum import SequenceIdEnums
 from slm_api.enums.fast_log_intervals_enum import FastLogInterval
+import datetime
 
 # ---------- Configuration ---------- 
 host = webxi_helper.set_host(__file__)
@@ -52,8 +53,8 @@ FAST_LOGGING_INTERVAL_MS = FastLogInterval.interval_1000ms.value
 SEQUENCE_ID = SequenceIdEnums.FastLAF.value
 
 class PrintDataHandler(DataHandler):
-    def handle(self, timestamp, name, local, value, unit):
-        print(f"{timestamp}{name} ({local}):  {value:7.2f} {unit}")
+    def handle(self, timestamp, name, local, value, unit, **data):
+        print(f"{timestamp} {name} ({local}):  {value:7.2f} {unit}")
 
 
 
@@ -150,14 +151,28 @@ class PrintDataHandler(DataHandler):
 
 
 if __name__ == "__main__":
+
+    filename = f'name-of-file_{datetime.datetime.now().strftime("%H%M_%m%d%Y")}'
+
+
+    # turns off all BB freq weights to prevent interference
+    webxi_helper.turn_off_bb_freq_weight(host)
+
+    # turns on the wanted BB freq weights for this example
+    webxi_helper.turn_on_bb_freq_weight(host, ["A"], fast=True)
+
+    # sets the sequences to true. 
+    # For fastlogging add fast=True so it routes to the correct endpoint
+    webxi_helper.turn_on_bb_leq(host, "LAF", fast=True)
     
     try:
         streamer = WebXiStreamHandler(host,
                                       sequenceID = SEQUENCE_ID, 
                                       mode="fast_logging",
                                       fast_logging_interval=FAST_LOGGING_INTERVAL_MS,
-                                      saving="csv",
-                                      saving_path=r"C:\Users\SKREBSVINGAA\csv_test\csv_test.txt")
+                                      saving="json",
+                                      saving_path=f"./saved_data/{filename}",
+                                      time=True)
         streamer.setDataHandler(PrintDataHandler())
                                                  
 

{slm_open_api_package-0.2.10 → slm_open_api_package-0.2.12}/src/slm_api/flac_example.py

@@ -29,6 +29,7 @@ import threading
 
 from slm_api.helpers.FigureHandler import FigureHandler
 from slm_api.helpers.data_handler import  MultiDataHandler, CsvDataHandler
+import datetime
 
 
 # FLAC streaming is only available on 2255
@@ -153,8 +154,10 @@ def on_close():
     streamer.stopStream()
 
 if __name__ == "__main__":
+    filename = f'name-of-file_{datetime.datetime.now().strftime("%H%M_%m%d%Y")}'
+
     # streamer = streamHandler()
-    streamer = sh.WebXiStreamHandler(host, sequenceID=sequenceID, mode="flac", saving="csv", saving_path=r"C:\Users\SKREBSVINGAA\csv_test\csv_test.txt")
+    streamer = sh.WebXiStreamHandler(host, sequenceID=sequenceID, mode="flac", saving="wav", saving_path=f"./saved_data/flac_{filename}")
     # streamer.setDataHandler(MultiDataHandler(PrintDataHandler(), CsvDataHandler(r"C:\Users\SKREBSVINGAA\csv_test\csv_test.txt")))
     fig = figureHandler()
     fig.app.aboutToQuit.connect(on_close)

slm_open_api_package-0.2.12/src/slm_api/helpers/data_handler.py

@@ -0,0 +1,251 @@
+"""Optional DataHandler for WebXiStreamHandler. By default WebXiStreamHandler
+does nothing with decoded values - call streamer.setDataHandler(...) with an
+instance of one of these (or your own subclass) before startStream() to
+receive/print/save them."""
+
+import csv
+import json
+import tempfile
+import threading
+import warnings
+from contextlib import contextmanager
+from pathlib import Path
+import numpy as np
+import soundfile as sf
+
+
+class DataHandler:
+    """Override handle() to decide what happens with decoded stream values.
+    """
+    
+
+    def handle(self, **data):
+        pass
+
+
+class _BufferedFileDataHandler(DataHandler):
+    """Buffers records and publishes snapshots in a background worker.
+    Call close() to stop the worker and save remaining records. WAV writes chunks.
+    """
+
+    _extension = None  # set by subclasses, e.g. "json", "csv", "wav"
+
+    def __init__(self, path, snapshot_interval=1.0):
+        if not np.isfinite(snapshot_interval) or snapshot_interval <= 0:
+            raise ValueError("Snapshot interval must be positive and finite")
+        path = Path(path)
+        if self._extension and path.suffix.lower() != f".{self._extension}":
+            path = path.with_name(path.name + f".{self._extension}")
+        path.parent.mkdir(parents=True, exist_ok=True)
+        self._path = path
+        self._records = []
+        self._lock = threading.RLock()
+        self._write_lock = threading.Lock()
+        self._snapshot_interval = snapshot_interval
+        self._stop_event = threading.Event()
+        self._worker = None
+        self._saved_count = -1
+        self._closed = False
+        self._write_error = None
+
+    def handle(self, **data):
+        with self._lock:
+            if self._closed:
+                return
+            if self._write_error is not None:
+                raise RuntimeError("Background snapshot saving failed") from self._write_error
+            if "values" in data:
+                data["values"] = [float(v) for v in data["values"]]
+            elif "samples" in data:
+                samples = data.pop("samples")
+                data.update(count=len(samples), min=float(min(samples)), max=float(max(samples)))
+            self._records.append(data)
+            if self._worker is None:
+                self._worker = threading.Thread(
+                    target=self._run_snapshots, name=f"snapshot-{self._path.name}", daemon=True,
+                )
+                self._worker.start()
+
+    def _run_snapshots(self):
+        while not self._stop_event.wait(self._snapshot_interval):
+            try:
+                self._flush_snapshot()
+            except Exception as error:
+                with self._lock:
+                    self._write_error = error
+                self._stop_event.set()
+                return
+
+    def _flush_snapshot(self):
+        with self._write_lock:
+            with self._lock:
+                count = len(self._records)
+                if count == self._saved_count:
+                    return
+                records = self._records.copy()
+            self._write_snapshot(records)
+            with self._lock:
+                self._saved_count = count
+
+    def _write_snapshot(self, records):
+        raise NotImplementedError
+
+    @contextmanager
+    def _snapshot_file(self, mode, **options):
+        temporary_path = None
+        try:
+            with tempfile.NamedTemporaryFile(
+                mode=mode, dir=self._path.parent, prefix=f".{self._path.name}.",
+                suffix=".tmp", delete=False, **options,
+            ) as target:
+                temporary_path = Path(target.name)
+                yield target
+            temporary_path.replace(self._path)
+        finally:
+            if temporary_path is not None:
+                temporary_path.unlink(missing_ok=True)
+
+    def close(self):
+        with self._lock:
+            self._closed = True
+            self._stop_event.set()
+            worker = self._worker
+        if worker is not None:
+            worker.join()
+        self._flush_snapshot()
+
+    def _group_records(self, records=None):
+        hosts = {}
+        for record in self._records if records is None else records:
+            row = record.copy()
+            if not row.get("timestamp"):
+                row.pop("timestamp", None)
+            host = row.pop("host", None)
+            rows = hosts.setdefault(host, [])
+            if "name" in row and "value" in row:
+                name = row.pop("name")
+                value = row.pop("value")
+                if (rows and rows[-1].get("timestamp") == row.get("timestamp")
+                        and name not in rows[-1]):
+                    rows[-1].update(row)
+                    rows[-1][name] = value
+                else:
+                    row[name] = value
+                    rows.append(row)
+            else:
+                rows.append(row)
+        return hosts
+
+
+class JsonDataHandler(_BufferedFileDataHandler):
+    """Writes records grouped by host, combining named values per timestamp."""
+    _extension = "json"
+
+    def _write_snapshot(self, records):
+        with self._snapshot_file("w", encoding="utf-8") as target:
+            records = [{"host": host, "data": rows} for host, rows in self._group_records(records).items()]
+            json.dump(
+                records,
+                target,
+                indent=2,
+                default=lambda value: (
+                    value.tolist()
+                    if isinstance(value, (np.ndarray, np.generic))
+                    else str(value)
+                ),
+            )
+
+class CsvDataHandler(_BufferedFileDataHandler):
+    """Writes rows per host and timestamp, storing vectors as JSON list cells."""
+    _extension = "csv"
+
+    def _write_snapshot(self, records):
+        with self._snapshot_file("w", encoding="utf-8", newline="") as target:
+            records = [
+                {"host": host, **row}
+                for host, rows in self._group_records(records).items()
+                for row in rows
+            ]
+            for record in records:
+                for name, value in record.items():
+                    if isinstance(value, np.ndarray):
+                        record[name] = json.dumps(value.tolist())
+                    elif isinstance(value, (list, tuple)):
+                        record[name] = json.dumps(value)
+            fields = dict.fromkeys(key for record in records for key in record)
+            fieldnames = [key for key in ("host", "timestamp") if key in fields]
+            fieldnames.extend(key for key in fields if key not in fieldnames)
+            writer = csv.DictWriter(target, fieldnames=fieldnames)
+            writer.writeheader()
+            writer.writerows(records)
+
+
+class WavDataHandler(_BufferedFileDataHandler):
+    """Writes PCM_24 chunks scaled by the full-scale pressure in the filename."""
+    _extension = "wav"
+
+    def __init__(self, path, sample_rate=None, full_scale_pa=None):
+        super().__init__(path)
+        self._sample_rate = sample_rate
+        self._full_scale_pa = full_scale_pa
+        self._started = False
+        self._closed = False
+
+    def handle(self, *, samples, sample_rate=None, full_scale_pa=None, **data):
+        with self._lock:
+            if self._closed:
+                return
+            rate = sample_rate if sample_rate is not None else self._sample_rate
+            if rate is None or rate <= 0 or int(rate) != rate:
+                raise ValueError("WAV samples require a positive integer sample rate")
+            if self._sample_rate is not None and rate != self._sample_rate:
+                raise ValueError("Sample rate cannot change within a WAV file")
+            scale = full_scale_pa if full_scale_pa is not None else self._full_scale_pa
+            if scale is None or not np.isfinite(scale) or scale <= 0:
+                raise ValueError("PCM_24 WAV samples require a positive finite full_scale_pa")
+            if self._full_scale_pa is not None and scale != self._full_scale_pa:
+                raise ValueError("Full-scale pressure cannot change within a WAV file")
+            values = np.asarray(samples, dtype=np.float64)
+            if values.ndim != 1:
+                raise ValueError("WAV samples must be a mono, one-dimensional array")
+            if not np.isfinite(values).all():
+                raise ValueError("WAV samples must be finite")
+            self._sample_rate = int(rate)
+            self._full_scale_pa = float(scale)
+            if values.size:
+                normalized = values / self._full_scale_pa
+                if np.any(normalized > 1.0) or np.any(normalized < -(2 ** 23) / (2 ** 23 - 1)):
+                    warnings.warn("Audio exceeds the calibrated full-scale pressure; PCM_24 samples will clip", RuntimeWarning, stacklevel=2)
+                normalized = np.clip(normalized, -1.0, 1.0)
+                if not self._started:
+                    self._path = self._path.with_name(
+                        f"{self._path.stem}_fullscale-{self._full_scale_pa:.17g}Pa.wav"
+                    )
+                options = {"mode": "r+"} if self._started else {
+                    "mode": "w", "samplerate": self._sample_rate,
+                    "channels": 1, "format": "WAV", "subtype": "PCM_24",
+                }
+                with sf.SoundFile(self._path, **options) as target:
+                    target.seek(0, sf.SEEK_END)
+                    target.write(normalized)
+                self._started = True
+
+    def close(self):
+        with self._lock:
+            self._closed = True
+
+
+class MultiDataHandler(DataHandler):
+    """Fans out to several handlers at once (e.g. print AND save AND plot)."""
+
+    def __init__(self, *handlers):
+        self._handlers = handlers
+
+    def handle(self, **data):
+        for h in self._handlers:
+            h.handle(**data)
+
+    def close(self):
+        for h in self._handlers:
+            if hasattr(h, "close"):
+                h.close()