From 48c8578b3224027b2d26b0db25e2f9793f287846 Mon Sep 17 00:00:00 2001
From: Gautzilla <gauthier.berthomieu@gmail.com>
Date: Mon, 16 Feb 2026 16:36:24 +0100
Subject: [PATCH 1/8] add seek methods

---
 src/osekit/audio_backend/audio_backend.py      |  2 ++
 src/osekit/audio_backend/audio_file_manager.py |  3 +++
 src/osekit/audio_backend/mseed_backend.py      |  3 +++
 src/osekit/audio_backend/soundfile_backend.py  |  7 +++++--
 src/osekit/core_api/audio_file.py              |  3 +++
 src/osekit/core_api/audio_item.py              | 12 ++++++++++++
 6 files changed, 28 insertions(+), 2 deletions(-)

diff --git a/src/osekit/audio_backend/audio_backend.py b/src/osekit/audio_backend/audio_backend.py
index f130ba24..599aaf89 100644
--- a/src/osekit/audio_backend/audio_backend.py
+++ b/src/osekit/audio_backend/audio_backend.py
@@ -47,3 +47,5 @@ def read(self, path: PathLike | str, start: int, stop: int) -> np.ndarray:
     def close(self) -> None:
         """Close the currently opened file."""
         ...
+
+    def seek(self, path: PathLike, frame: int) -> None: ...
diff --git a/src/osekit/audio_backend/audio_file_manager.py b/src/osekit/audio_backend/audio_file_manager.py
index 7ff42de6..b1df4591 100644
--- a/src/osekit/audio_backend/audio_file_manager.py
+++ b/src/osekit/audio_backend/audio_file_manager.py
@@ -102,3 +102,6 @@ def read(
             raise ValueError(msg)
 
         return self._backend(path).read(path=path, start=start, stop=stop)
+
+    def seek(self, path: Path, frame: int) -> None:
+        self._backend(path=path).seek(path=path, frame=frame)
diff --git a/src/osekit/audio_backend/mseed_backend.py b/src/osekit/audio_backend/mseed_backend.py
index d9e86e4e..6472eaac 100644
--- a/src/osekit/audio_backend/mseed_backend.py
+++ b/src/osekit/audio_backend/mseed_backend.py
@@ -83,3 +83,6 @@ def read(
 
         data = np.concatenate([trace.data for trace in file_content])
         return data[start:stop]
+
+    def seek(self, path: PathLike, frame: int) -> None:
+        pass
diff --git a/src/osekit/audio_backend/soundfile_backend.py b/src/osekit/audio_backend/soundfile_backend.py
index 11cc36ea..946ce0d2 100644
--- a/src/osekit/audio_backend/soundfile_backend.py
+++ b/src/osekit/audio_backend/soundfile_backend.py
@@ -65,10 +65,13 @@ def read(
             A ``(channel * frames)`` array containing the audio data.
 
         """
-        self._switch(path)
-        self._file.seek(start)
+        self.seek(path=path, frame=start)
         return self._file.read(stop - start)
 
+    def seek(self, path: PathLike, frame: int) -> None:
+        self._switch(path=path)
+        self._file.seek(frame)
+
     def _close(self) -> None:
         if self._file is None:
             return
diff --git a/src/osekit/core_api/audio_file.py b/src/osekit/core_api/audio_file.py
index f825d376..f89daf48 100644
--- a/src/osekit/core_api/audio_file.py
+++ b/src/osekit/core_api/audio_file.py
@@ -132,3 +132,6 @@ def move(self, folder: Path) -> None:
         """
         afm.close()
         super().move(folder)
+
+    def seek(self, frame: int) -> None:
+        afm.seek(path=self.path, frame=frame)
diff --git a/src/osekit/core_api/audio_item.py b/src/osekit/core_api/audio_item.py
index 20fdacea..77612323 100644
--- a/src/osekit/core_api/audio_item.py
+++ b/src/osekit/core_api/audio_item.py
@@ -64,3 +64,15 @@ def get_value(self) -> np.ndarray:
         if self.is_empty:
             return np.zeros((1, self.nb_channels))
         return super().get_value()
+
+    def stream(self, chunk_size: int) -> np.ndarray:
+        start_frame, stop_frame = self.file.frames_indexes(
+            start=self.begin, stop=self.end
+        )
+
+        remaining = stop_frame - start_frame
+
+        self.file.seek(frame=start_frame)
+
+        while remaining > 0:
+            pass

From 349f8a0cea7b2d5259d85617b300aac23881cc12 Mon Sep 17 00:00:00 2001
From: Gautzilla <gauthier.berthomieu@gmail.com>
Date: Mon, 16 Feb 2026 18:22:19 +0100
Subject: [PATCH 2/8] add stream(() methods

---
 src/osekit/audio_backend/audio_backend.py      |  2 ++
 src/osekit/audio_backend/audio_file_manager.py |  3 +++
 src/osekit/audio_backend/mseed_backend.py      |  3 +++
 src/osekit/audio_backend/soundfile_backend.py  |  4 ++++
 src/osekit/core_api/audio_file.py              |  3 +++
 src/osekit/core_api/audio_item.py              | 10 +++++++---
 6 files changed, 22 insertions(+), 3 deletions(-)

diff --git a/src/osekit/audio_backend/audio_backend.py b/src/osekit/audio_backend/audio_backend.py
index 599aaf89..9eadc890 100644
--- a/src/osekit/audio_backend/audio_backend.py
+++ b/src/osekit/audio_backend/audio_backend.py
@@ -49,3 +49,5 @@ def close(self) -> None:
         ...
 
     def seek(self, path: PathLike, frame: int) -> None: ...
+
+    def stream(self, path: PathLike, chunk_size: int) -> np.ndarray: ...
diff --git a/src/osekit/audio_backend/audio_file_manager.py b/src/osekit/audio_backend/audio_file_manager.py
index b1df4591..a3d836e4 100644
--- a/src/osekit/audio_backend/audio_file_manager.py
+++ b/src/osekit/audio_backend/audio_file_manager.py
@@ -105,3 +105,6 @@ def read(
 
     def seek(self, path: Path, frame: int) -> None:
         self._backend(path=path).seek(path=path, frame=frame)
+
+    def stream(self, path: Path, chunk_size: int) -> np.ndarray:
+        self._backend(path=path).stream(path=path, chunk_size=chunk_size)
diff --git a/src/osekit/audio_backend/mseed_backend.py b/src/osekit/audio_backend/mseed_backend.py
index 6472eaac..b1218f6e 100644
--- a/src/osekit/audio_backend/mseed_backend.py
+++ b/src/osekit/audio_backend/mseed_backend.py
@@ -86,3 +86,6 @@ def read(
 
     def seek(self, path: PathLike, frame: int) -> None:
         pass
+
+    def stream(self, path: PathLike, chunk_size: int) -> np.ndarray:
+        pass
diff --git a/src/osekit/audio_backend/soundfile_backend.py b/src/osekit/audio_backend/soundfile_backend.py
index 946ce0d2..7bce1119 100644
--- a/src/osekit/audio_backend/soundfile_backend.py
+++ b/src/osekit/audio_backend/soundfile_backend.py
@@ -72,6 +72,10 @@ def seek(self, path: PathLike, frame: int) -> None:
         self._switch(path=path)
         self._file.seek(frame)
 
+    def stream(self, path: PathLike, chunk_size: int) -> np.ndarray:
+        self._switch(path=path)
+        return self._file.read(frames=chunk_size)
+
     def _close(self) -> None:
         if self._file is None:
             return
diff --git a/src/osekit/core_api/audio_file.py b/src/osekit/core_api/audio_file.py
index f89daf48..bd97d95f 100644
--- a/src/osekit/core_api/audio_file.py
+++ b/src/osekit/core_api/audio_file.py
@@ -135,3 +135,6 @@ def move(self, folder: Path) -> None:
 
     def seek(self, frame: int) -> None:
         afm.seek(path=self.path, frame=frame)
+
+    def stream(self, chunk_size: int) -> np.ndarray:
+        return afm.stream(path=self.path, chunk_size=chunk_size)
diff --git a/src/osekit/core_api/audio_item.py b/src/osekit/core_api/audio_item.py
index 77612323..ef5bfcbb 100644
--- a/src/osekit/core_api/audio_item.py
+++ b/src/osekit/core_api/audio_item.py
@@ -2,6 +2,7 @@
 
 from __future__ import annotations
 
+from collections.abc import Generator
 from typing import TYPE_CHECKING
 
 import numpy as np
@@ -65,9 +66,10 @@ def get_value(self) -> np.ndarray:
             return np.zeros((1, self.nb_channels))
         return super().get_value()
 
-    def stream(self, chunk_size: int) -> np.ndarray:
+    def stream(self, chunk_size: int) -> Generator[np.ndarray, None, None]:
         start_frame, stop_frame = self.file.frames_indexes(
-            start=self.begin, stop=self.end
+            start=self.begin,
+            stop=self.end,
         )
 
         remaining = stop_frame - start_frame
@@ -75,4 +77,6 @@ def stream(self, chunk_size: int) -> np.ndarray:
         self.file.seek(frame=start_frame)
 
         while remaining > 0:
-            pass
+            frames_to_read = min(chunk_size, remaining)
+            yield self.file.stream(chunk_size=frames_to_read)
+            remaining -= frames_to_read

From 7c90852d7dcc9f261ef30151b22b3ea7dfc141f0 Mon Sep 17 00:00:00 2001
From: Gautzilla <gauthier.berthomieu@gmail.com>
Date: Tue, 17 Feb 2026 16:35:06 +0100
Subject: [PATCH 3/8] add soxr streamed resample

---
 .../audio_backend/audio_file_manager.py       |  2 +-
 src/osekit/core_api/audio_data.py             | 56 +++++++++++++++++++
 src/osekit/core_api/audio_file.py             | 12 ++--
 3 files changed, 63 insertions(+), 7 deletions(-)

diff --git a/src/osekit/audio_backend/audio_file_manager.py b/src/osekit/audio_backend/audio_file_manager.py
index a3d836e4..da0bbdbb 100644
--- a/src/osekit/audio_backend/audio_file_manager.py
+++ b/src/osekit/audio_backend/audio_file_manager.py
@@ -107,4 +107,4 @@ def seek(self, path: Path, frame: int) -> None:
         self._backend(path=path).seek(path=path, frame=frame)
 
     def stream(self, path: Path, chunk_size: int) -> np.ndarray:
-        self._backend(path=path).stream(path=path, chunk_size=chunk_size)
+        return self._backend(path=path).stream(path=path, chunk_size=chunk_size)
diff --git a/src/osekit/core_api/audio_data.py b/src/osekit/core_api/audio_data.py
index 081d08b2..55101a42 100644
--- a/src/osekit/core_api/audio_data.py
+++ b/src/osekit/core_api/audio_data.py
@@ -6,13 +6,16 @@
 
 from __future__ import annotations
 
+from collections.abc import Generator
 from math import ceil
 from typing import TYPE_CHECKING, Self
 
 import numpy as np
 import soundfile as sf
+import soxr
 from pandas import Timedelta, Timestamp
 
+from osekit.config import resample_quality_settings
 from osekit.core_api.audio_file import AudioFile
 from osekit.core_api.audio_item import AudioItem
 from osekit.core_api.base_data import BaseData
@@ -216,6 +219,7 @@ def get_raw_value(self) -> np.ndarray:
             The value of the audio data.
 
         """
+        return np.vstack(list(self.stream()))
         data = np.empty(shape=self.shape)
         idx = 0
         for item in self.items:
@@ -225,6 +229,58 @@ def get_raw_value(self) -> np.ndarray:
             idx += len(item_data)
         return data
 
+    def stream(self, chunk_size: int = 8192) -> Generator[np.ndarray, None, None]:
+        resampler = None
+        input_sr = None
+        produced_samples = 0
+        total_samples = self.length
+
+        for item in self.items:
+            if item.is_empty:
+                silence_length = round(item.duration.total_seconds() * self.sample_rate)
+                yield item.get_value().repeat(
+                    silence_length,
+                    axis=0,
+                )
+                produced_samples += silence_length
+                continue
+
+            if (resampler is None) or (input_sr != item.sample_rate):
+                input_sr = item.sample_rate
+                quality = resample_quality_settings[
+                    "downsample" if input_sr > self.sample_rate else "upsample"
+                ]
+                resampler = soxr.ResampleStream(
+                    in_rate=input_sr,
+                    out_rate=self.sample_rate,
+                    num_channels=self.nb_channels,
+                    quality=quality,
+                    dtype=np.float64,
+                )
+
+            for chunk in item.stream(chunk_size=chunk_size):
+                y = chunk
+                if item.sample_rate != self.sample_rate:
+                    y = resampler.resample_chunk(x=chunk)
+
+                remaining = total_samples - produced_samples
+                y = y[:remaining]
+                produced_samples += len(y)
+
+                yield y
+
+                if produced_samples >= total_samples:
+                    return
+
+        if resampler is None:
+            return
+
+        flush = resampler.resample_chunk(np.array([]), last=True)
+        if len(flush) and (remaining := (total_samples - produced_samples)):
+            if flush.ndim == 1:
+                flush = flush[:, None]
+            yield flush[:remaining]
+
     def get_value(self) -> np.ndarray:
         """Return the value of the audio data.
 
diff --git a/src/osekit/core_api/audio_file.py b/src/osekit/core_api/audio_file.py
index bd97d95f..0bfea239 100644
--- a/src/osekit/core_api/audio_file.py
+++ b/src/osekit/core_api/audio_file.py
@@ -89,11 +89,8 @@ def read(self, start: Timestamp, stop: Timestamp) -> np.ndarray:
         """
         start_sample, stop_sample = self.frames_indexes(start, stop)
         data = afm.read(self.path, start=start_sample, stop=stop_sample)
-        if len(data.shape) == 1:
-            return data.reshape(
-                data.shape[0],
-                1,
-            )  # 2D array to match the format of multichannel audio
+        if data.ndim == 1:
+            return data[:, None]  # 2D array to match the format of multichannel audio
         return data
 
     def frames_indexes(self, start: Timestamp, stop: Timestamp) -> tuple[int, int]:
@@ -137,4 +134,7 @@ def seek(self, frame: int) -> None:
         afm.seek(path=self.path, frame=frame)
 
     def stream(self, chunk_size: int) -> np.ndarray:
-        return afm.stream(path=self.path, chunk_size=chunk_size)
+        data = afm.stream(path=self.path, chunk_size=chunk_size)
+        if data.ndim == 1:
+            return data[:, None]  # 2D array to match the format of multichannel audio
+        return data

From 0a4dc5b745ca56c847d153e13533ac1c99b739c3 Mon Sep 17 00:00:00 2001
From: Gautzilla <gauthier.berthomieu@gmail.com>
Date: Wed, 18 Feb 2026 17:18:00 +0100
Subject: [PATCH 4/8] =?UTF-8?q?add=20stream()=20and=20seek(=C3=83)=20place?=
 =?UTF-8?q?holders=20for=20MSEED=20backend?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 src/osekit/audio_backend/mseed_backend.py | 33 +++++++++++++++++++++--
 1 file changed, 31 insertions(+), 2 deletions(-)

diff --git a/src/osekit/audio_backend/mseed_backend.py b/src/osekit/audio_backend/mseed_backend.py
index b1218f6e..e15ac5e0 100644
--- a/src/osekit/audio_backend/mseed_backend.py
+++ b/src/osekit/audio_backend/mseed_backend.py
@@ -19,6 +19,7 @@ class MSeedBackend:
     def __init__(self) -> None:
         """Initialize the MSEED backend."""
         _require_obspy()
+        self.seeked_frame = 0
 
     def close(self) -> None:
         """Close the currently opened file. No use in MSEED files."""
@@ -85,7 +86,35 @@ def read(
         return data[start:stop]
 
     def seek(self, path: PathLike, frame: int) -> None:
-        pass
+        """Set the seeked_frame of the backend.
+
+        Streamed data will be streamed from this frame.
+
+        Parameters
+        ----------
+        path: PathLike | str
+            No effect.
+        frame: int
+            Frame to seek.
+
+        """
+        self.seeked_frame = frame
 
     def stream(self, path: PathLike, chunk_size: int) -> np.ndarray:
-        pass
+        """Stream the content of the MSEED file from the seeked frame.
+
+        Parameters
+        ----------
+        path: PathLike
+            Path to the mseed file.
+        chunk_size: int
+            Number of frames to stream.
+
+        Returns
+        -------
+        np.ndarray:
+            Streamed data of length ``chunk_size`` from ``self.seeked_frame``.
+        """
+        return self.read(
+            path=path, start=self.seeked_frame, stop=self.seeked_frame + chunk_size
+        )

From 4316a70e61e0478624613d9de7cba3be18b87fba Mon Sep 17 00:00:00 2001
From: Gautzilla <gauthier.berthomieu@gmail.com>
Date: Wed, 18 Feb 2026 17:47:03 +0100
Subject: [PATCH 5/8] add docstrings in seek and stream functions

---
 src/osekit/core_api/audio_data.py | 36 +++++++++++++------------------
 src/osekit/core_api/audio_file.py | 21 ++++++++++++++++++
 2 files changed, 36 insertions(+), 21 deletions(-)

diff --git a/src/osekit/core_api/audio_data.py b/src/osekit/core_api/audio_data.py
index 55101a42..2b4b063f 100644
--- a/src/osekit/core_api/audio_data.py
+++ b/src/osekit/core_api/audio_data.py
@@ -20,7 +20,7 @@
 from osekit.core_api.audio_item import AudioItem
 from osekit.core_api.base_data import BaseData
 from osekit.core_api.instrument import Instrument
-from osekit.utils.audio_utils import Normalization, normalize, resample
+from osekit.utils.audio_utils import Normalization, normalize
 
 if TYPE_CHECKING:
     from pathlib import Path
@@ -220,16 +220,22 @@ def get_raw_value(self) -> np.ndarray:
 
         """
         return np.vstack(list(self.stream()))
-        data = np.empty(shape=self.shape)
-        idx = 0
-        for item in self.items:
-            item_data = self._get_item_value(item)
-            item_data = item_data[: min(item_data.shape[0], data.shape[0] - idx)]
-            data[idx : idx + len(item_data)] = item_data
-            idx += len(item_data)
-        return data
 
     def stream(self, chunk_size: int = 8192) -> Generator[np.ndarray, None, None]:
+        """Stream the audio data in chunks.
+
+        Parameters
+        ----------
+        chunk_size: int
+            Size of the chunks of audio yielded by the generator.
+
+        Returns
+        -------
+        Generator[np.ndarray, None, None]:
+            Generated ``np.ndarray`` of dimensions (``chunk_size``*``self.nb_channels``)
+            of the streamed audio data.
+
+        """
         resampler = None
         input_sr = None
         produced_samples = 0
@@ -369,18 +375,6 @@ def link(self, folder: Path) -> None:
         )
         self.items = AudioData.from_files([file]).items
 
-    def _get_item_value(self, item: AudioItem) -> np.ndarray:
-        """Return the resampled (if needed) data from the audio item."""
-        item_data = item.get_value()
-        if item.is_empty:
-            return item_data.repeat(
-                round(item.duration.total_seconds() * self.sample_rate),
-                axis=0,
-            )
-        if item.sample_rate != self.sample_rate:
-            return resample(item_data, item.sample_rate, self.sample_rate)
-        return item_data
-
     def split(
         self,
         nb_subdata: int = 2,
diff --git a/src/osekit/core_api/audio_file.py b/src/osekit/core_api/audio_file.py
index 0bfea239..f52c15c6 100644
--- a/src/osekit/core_api/audio_file.py
+++ b/src/osekit/core_api/audio_file.py
@@ -131,9 +131,30 @@ def move(self, folder: Path) -> None:
         super().move(folder)
 
     def seek(self, frame: int) -> None:
+        """Seek the requested frame in the file.
+
+        Parameters
+        ----------
+        frame: int
+            Index of the frame to be seeked.
+
+        """
         afm.seek(path=self.path, frame=frame)
 
     def stream(self, chunk_size: int) -> np.ndarray:
+        """Stream ``chunk_size`` frames from the audio file.
+
+        Parameters
+        ----------
+        chunk_size: int
+            Number of frames to stream from the audio file.
+
+        Returns
+        -------
+        np.ndarray:
+            A (``chunk_size``*``self.channels``) array of frames.
+
+        """
         data = afm.stream(path=self.path, chunk_size=chunk_size)
         if data.ndim == 1:
             return data[:, None]  # 2D array to match the format of multichannel audio

From 9b1597d1f7ae6fffc7e611d49aca3c6f3959309b Mon Sep 17 00:00:00 2001
From: Gautzilla <gauthier.berthomieu@gmail.com>
Date: Wed, 18 Feb 2026 17:56:52 +0100
Subject: [PATCH 6/8] add tests for AudioFile stream shape

---
 tests/test_audio.py | 26 ++++++++++++++++++++++++++
 1 file changed, 26 insertions(+)

diff --git a/tests/test_audio.py b/tests/test_audio.py
index 12e4f049..b3a494c2 100644
--- a/tests/test_audio.py
+++ b/tests/test_audio.py
@@ -222,6 +222,32 @@ def test_audio_file_read(
     assert np.allclose(files[0].read(start, stop)[:, 0], expected, atol=1e-7)
 
 
+@pytest.mark.parametrize(
+    ("mocked_data", "expected_shape"),
+    [
+        pytest.param(np.array([0, 1, 2, 3]), (4, 1), id="1d-to-2d"),
+        pytest.param(np.array([[0, 1], [2, 3], [4, 5], [6, 7]]), (4, 2), id="2d-to-2d"),
+    ],
+)
+def test_audio_file_stream_is_always_2d(
+    monkeypatch: pytest.MonkeyPatch, mocked_data: np.ndarray, expected_shape: tuple
+) -> None:
+    def mocked_stream(*args: None, **kwargs: None) -> np.ndarray:
+        return mocked_data
+
+    monkeypatch.setattr("osekit.core_api.audio_file.afm.stream", mocked_stream)
+
+    def mocked_init(self: AudioFile, *args: None, **kwargs: None) -> None:
+        self.path = Path()
+        self.begin = Timestamp("1996-04-15 00:00:00")
+
+    monkeypatch.setattr(AudioFile, "__init__", mocked_init)
+
+    af = AudioFile()
+
+    assert af.stream(1024).shape == expected_shape
+
+
 def test_multichannel_audio_file_read(monkeypatch: pytest.MonkeyPatch) -> None:
     full_file = np.array([[1, 1, 1], [2, 2, 2], [3, 3, 3], [4, 4, 4], [5, 5, 5]])
 

From 78dbd00105c162bca4ef14d99ba7dac37a8f0112 Mon Sep 17 00:00:00 2001
From: Gautzilla <gauthier.berthomieu@gmail.com>
Date: Thu, 19 Feb 2026 11:19:10 +0100
Subject: [PATCH 7/8] add resample from multiple sample rates test

---
 tests/test_audio.py | 25 ++++++++++++++++++++++++-
 1 file changed, 24 insertions(+), 1 deletion(-)

diff --git a/tests/test_audio.py b/tests/test_audio.py
index b3a494c2..71659d05 100644
--- a/tests/test_audio.py
+++ b/tests/test_audio.py
@@ -230,7 +230,9 @@ def test_audio_file_read(
     ],
 )
 def test_audio_file_stream_is_always_2d(
-    monkeypatch: pytest.MonkeyPatch, mocked_data: np.ndarray, expected_shape: tuple
+    monkeypatch: pytest.MonkeyPatch,
+    mocked_data: np.ndarray,
+    expected_shape: tuple,
 ) -> None:
     def mocked_stream(*args: None, **kwargs: None) -> np.ndarray:
         return mocked_data
@@ -742,6 +744,27 @@ def test_audio_resample_sample_count(
     assert data.get_value().shape[0] == expected_nb_samples
 
 
+def test_audio_resample_different_samplerates(tmp_path: Path) -> None:
+    fs1 = 20
+    fs2 = 10
+    d1 = 1
+    d2 = 1
+    s1 = np.linspace(-1.0, 0.0, fs1 * d1)
+    s2 = np.linspace(0.0, 1.0, fs2 * d2)
+
+    p1 = tmp_path / "s1.wav"
+    p2 = tmp_path / "s2.wav"
+
+    sf.write(file=p1, data=s1, samplerate=fs1)
+    sf.write(file=p2, data=s2, samplerate=fs2)
+
+    af1 = AudioFile(path=p1, begin=Timestamp("2020-01-01 00:00:00"))
+    af2 = AudioFile(path=p2, begin=Timestamp("2020-01-01 00:00:01"))
+
+    ad = AudioData.from_files(files=[af1, af2], sample_rate=10)
+    assert ad.get_value().shape == (20, 1)
+
+
 @pytest.mark.parametrize(
     ("audio_files", "downsampling_quality", "upsampling_quality"),
     [

From 12fe6be46f33dce6d20ae3610c30f48f0ee0973c Mon Sep 17 00:00:00 2001
From: Gautzilla <gauthier.berthomieu@gmail.com>
Date: Thu, 19 Feb 2026 11:19:39 +0100
Subject: [PATCH 8/8] flush resampler before switching to a new one

---
 src/osekit/core_api/audio_data.py | 29 ++++++++++++++++++++++++-----
 1 file changed, 24 insertions(+), 5 deletions(-)

diff --git a/src/osekit/core_api/audio_data.py b/src/osekit/core_api/audio_data.py
index 2b4b063f..98210010 100644
--- a/src/osekit/core_api/audio_data.py
+++ b/src/osekit/core_api/audio_data.py
@@ -221,6 +221,19 @@ def get_raw_value(self) -> np.ndarray:
         """
         return np.vstack(list(self.stream()))
 
+    @staticmethod
+    def _flush(
+        resampler: soxr.ResampleStream,
+        remaining_samples: int,
+    ) -> np.ndarray:
+        flush = resampler.resample_chunk(np.array([]), last=True)
+        if len(flush) == 0:
+            return np.array([])
+        if not remaining_samples:
+            return np.array([])
+        flush = flush[:remaining_samples]
+        return flush[:, None] if flush.ndim == 1 else flush
+
     def stream(self, chunk_size: int = 8192) -> Generator[np.ndarray, None, None]:
         """Stream the audio data in chunks.
 
@@ -252,6 +265,13 @@ def stream(self, chunk_size: int = 8192) -> Generator[np.ndarray, None, None]:
                 continue
 
             if (resampler is None) or (input_sr != item.sample_rate):
+                if resampler:
+                    flush = self._flush(
+                        resampler=resampler,
+                        remaining_samples=total_samples - produced_samples,
+                    )
+                    yield flush
+                    produced_samples += len(flush[0])
                 input_sr = item.sample_rate
                 quality = resample_quality_settings[
                     "downsample" if input_sr > self.sample_rate else "upsample"
@@ -281,11 +301,10 @@ def stream(self, chunk_size: int = 8192) -> Generator[np.ndarray, None, None]:
         if resampler is None:
             return
 
-        flush = resampler.resample_chunk(np.array([]), last=True)
-        if len(flush) and (remaining := (total_samples - produced_samples)):
-            if flush.ndim == 1:
-                flush = flush[:, None]
-            yield flush[:remaining]
+        yield self._flush(
+            resampler=resampler,
+            remaining_samples=total_samples - produced_samples,
+        )
 
     def get_value(self) -> np.ndarray:
         """Return the value of the audio data.