fix: correct error handling in process_reverse_stream

livekit-rtc/livekit/rtc/apm.py

@@ -90,8 +90,8 @@ def process_reverse_stream(self, data: AudioFrame) -> None:
 
         resp = FfiClient.instance.request(req)
 
-        if resp.apm_process_stream.error:
-            raise RuntimeError(resp.apm_process_stream.error)
+        if resp.apm_process_reverse_stream.error:
+            raise RuntimeError(resp.apm_process_reverse_stream.error)
 
     def set_stream_delay_ms(self, delay_ms: int) -> None:
         """

tests/rtc/test_agc.py

@@ -0,0 +1,60 @@
+import os
+import wave
+import numpy as np
+
+from livekit.rtc import AudioProcessingModule, AudioFrame
+
+FIXTURES_DIR = os.path.join(os.path.dirname(os.path.abspath(__file__)), "fixtures")
+
+
+def test_agc_modifies_audio():
+    num_channels = 1
+
+    input_wav = os.path.join(FIXTURES_DIR, "test_audio.wav")
+    output_wav = os.path.join(FIXTURES_DIR, "test_processed.wav")
+
+    apm = AudioProcessingModule(auto_gain_control=True)
+
+    any_frame_modified = False
+
+    with wave.open(input_wav, "rb") as wf_in:
+        assert wf_in.getnchannels() == num_channels, "Input file must be mono."
+        sample_rate = wf_in.getframerate()
+        sampwidth = wf_in.getsampwidth()
+        frames_per_chunk = sample_rate // 100
+
+        with wave.open(output_wav, "wb") as wf_out:
+            wf_out.setnchannels(num_channels)
+            wf_out.setsampwidth(sampwidth)
+            wf_out.setframerate(sample_rate)
+
+            while True:
+                raw_bytes = wf_in.readframes(frames_per_chunk)
+                if not raw_bytes:
+                    break
+
+                data = np.frombuffer(raw_bytes, dtype=np.int16)
+                if len(data) < frames_per_chunk:
+                    data = np.pad(data, (0, frames_per_chunk - len(data)))
+
+                original = data.copy()
+
+                frame = AudioFrame(
+                    data=data.tobytes(),
+                    sample_rate=sample_rate,
+                    num_channels=num_channels,
+                    samples_per_channel=frames_per_chunk,
+                )
+
+                apm.process_stream(frame)
+
+                processed = np.frombuffer(frame.data, dtype=np.int16)
+                if not np.array_equal(original, processed):
+                    any_frame_modified = True
+
+                wf_out.writeframes(frame.data.tobytes())
+
+    assert any_frame_modified, (
+        "APM did not modify any audio frames — processing may be a no-op. "
+        "With AGC enabled, output should differ from input."
+    )