diff --git a/gui/src/effects.rs b/gui/src/effects.rs
index e85c945..9f6c821 100644
--- a/gui/src/effects.rs
+++ b/gui/src/effects.rs
@@ -74,6 +74,7 @@ fn with_denoising_model<O>(
 pub struct AudioProcessor {
     denoise: bool,
     spawn: imp::SpawnHandle,
+    buffer: Vec<f32>,
 }
 
 impl AudioProcessor {
@@ -81,6 +82,7 @@ impl AudioProcessor {
         AudioProcessor {
             denoise: false,
             spawn: imp::SpawnHandle::current(),
+            buffer: Vec::new(),
         }
     }
 
@@ -88,30 +90,45 @@ impl AudioProcessor {
         AudioProcessor {
             denoise: true,
             spawn: imp::SpawnHandle::current(),
+            buffer: Vec::new(),
         }
     }
 }
 
 impl AudioProcessor {
     pub fn process(&mut self, audio: &[f32], output: &mut Vec<f32>) {
-        let mut finished = false;
+        let mut include_raw = true;
         if self.denoise {
             with_denoising_model(&self.spawn, |df| {
-                let start = output.len();
-                output.extend(std::iter::repeat_n(0f32, audio.len()));
-                finished = true;
-                let output = &mut output[start..];
-                df.process(
-                    slice_as_arrayview(audio, &[audio.len()])
-                        .into_shape((1, audio.len()))
-                        .unwrap(),
-                    mut_slice_as_arrayviewmut(output, &[output.len()])
-                        .into_shape((1, output.len()))
-                        .unwrap(),
-                );
+                include_raw = false;
+
+                self.buffer.extend_from_slice(audio);
+                output.reserve(audio.len());
+
+                let hop = df.hop_size;
+                let mut i = 0;
+                while self.buffer[i..].len() >= hop {
+                    let audio = &self.buffer[i..][..hop];
+                    i += audio.len();
+
+                    let j = output.len();
+                    output.extend(std::iter::repeat_n(0f32, audio.len()));
+                    let output = &mut output[j..];
+
+                    df.process(
+                        slice_as_arrayview(audio, &[audio.len()])
+                            .into_shape((1, audio.len()))
+                            .unwrap(),
+                        mut_slice_as_arrayviewmut(output, &[output.len()])
+                            .into_shape((1, output.len()))
+                            .unwrap(),
+                    );
+                }
+                self.buffer.splice(..i, []);
             });
         }
-        if !finished {
+
+        if include_raw {
             output.extend_from_slice(audio);
         }
     }