audio(filters): add noisereduce filter

CHANGELOG.md

@@ -9,6 +9,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Added
 
+- Added audio filter `NoisereduceFilter`.
+
 - Introduce input transport audio filters (`BaseAudioFilter`). Audio filters can
   be used to remove background noises before audio is sent to VAD.
 

pyproject.toml

@@ -56,6 +56,7 @@ livekit = [ "livekit~=0.17.5", "livekit-api~=0.7.1", "tenacity~=8.5.0" ]
 lmnt = [ "lmnt~=1.1.4" ]
 local = [ "pyaudio~=0.2.14" ]
 moondream = [ "einops~=0.8.0", "timm~=1.0.8", "transformers~=4.44.0" ]
+noisereduce = [ "noisereduce~=3.0.3" ]
 openai = [ "openai~=1.50.2", "websockets~=13.1", "python-deepcompare~=1.0.1" ]
 openpipe = [ "openpipe~=4.24.0" ]
 playht = [ "pyht~=0.1.4", "websockets~=13.1" ]

src/pipecat/audio/filters/noisereduce_filter.py

@@ -0,0 +1,49 @@
+#
+# Copyright (c) 2024, Daily
+#
+# SPDX-License-Identifier: BSD 2-Clause License
+#
+
+import numpy as np
+
+from pipecat.audio.filters.base_audio_filter import BaseAudioFilter
+
+from loguru import logger
+
+from pipecat.frames.frames import FilterControlFrame
+
+try:
+    import noisereduce as nr
+except ModuleNotFoundError as e:
+    logger.error(f"Exception: {e}")
+    logger.error(
+        "In order to use the noisereduce filter, you need to `pip install pipecat-ai[noisereduce]`."
+    )
+    raise Exception(f"Missing module: {e}")
+
+
+class NoisereduceFilter(BaseAudioFilter):
+    def __init__(self) -> None:
+        self._sample_rate = 0
+
+    async def start(self, sample_rate: int):
+        self._sample_rate = sample_rate
+
+    async def stop(self):
+        pass
+
+    async def process_frame(self, frame: FilterControlFrame):
+        pass
+
+    async def filter(self, audio: bytes) -> bytes:
+        data = np.frombuffer(audio, dtype=np.int16)
+
+        # Add a small epsilon to avoid division by zero.
+        epsilon = 1e-10
+        data = data.astype(np.float32) + epsilon
+
+        # Noise reduction
+        reduced_noise = nr.reduce_noise(y=data, sr=self._sample_rate)
+        audio = np.clip(reduced_noise, -32768, 32767).astype(np.int16).tobytes()
+
+        return audio