pipecat-ai
diff --git a/‎changelog/5667.added.md‎
Lines changed: 1 addition & 0 deletions b/‎changelog/5667.added.md‎
Lines changed: 1 addition & 0 deletions
diff --git a/‎uv.lock‎
Lines changed: 24 additions & 1 deletion b/‎uv.lock‎
Lines changed: 24 additions & 1 deletion
@@ -0,0 +1 @@
+- Added `DPDFNetFilter`, an input audio filter using [DPDFNet](https://github.com/ceva-ip/DPDFNet) for real-time speech enhancement on CPU via ONNX Runtime. Inference runs on a dedicated worker thread so the event loop is never blocked, and audio is resampled to and from the model's native rate with `SOXRStreamAudioResampler` when the transport rate differs. Select a model with `model=` to trade quality for CPU (`baseline` is cheapest, `dpdfnet2` is the default); 8 kHz and 48 kHz native variants avoid resampling on telephony and WebRTC transports. If the model cannot be loaded the filter passes audio through unmodified. Install with `uv add "pipecat-ai[dpdfnet]"`.
Original file line number	Diff line number	Diff line change
`@@ -0,0 +1 @@`
	`1`	+- Added `DPDFNetFilter`, an input audio filter using [DPDFNet](https://github.com/ceva-ip/DPDFNet) for real-time speech enhancement on CPU via ONNX Runtime. Inference runs on a dedicated worker thread so the event loop is never blocked, and audio is resampled to and from the model's native rate with `SOXRStreamAudioResampler` when the transport rate differs. Select a model with `model=` to trade quality for CPU (`baseline` is cheapest, `dpdfnet2` is the default); 8 kHz and 48 kHz native variants avoid resampling on telephony and WebRTC transports. If the model cannot be loaded the filter passes audio through unmodified. Install with `uv add "pipecat-ai[dpdfnet]"`.