diff --git a/docs/models/hey_mycroft.md b/docs/models/hey_mycroft.md
index 912fa72..69e094c 100644
--- a/docs/models/hey_mycroft.md
+++ b/docs/models/hey_mycroft.md
@@ -75,8 +75,8 @@ The positive test examples of the "hey mycroft" wakeword were collected manually
 
 The false-accept/false-reject curve for the model on the test data is shown below. Decreasing the `threshold` parameter when using the model will increase the false-accept rate and decrease the false-reject rate.
 
-![FPR/FRR curve for "hey mycroft" pre-trained model](images/hey_mycroft_performance_plot.png)
+![FPR/FRR curve for "hey mycroft" pre-trained model](images/hey_mycroft_performance.png)
 
 # Other Considerations
 
-While the model was trained to be robust to background noise and reverberation, it will still perform the best when the audio is relatively clean and free of overly loud background noise. In particular, the presence of audio playback of music/speech from the same device that is capturing the microphone stream may result in significantly higher false-reject rates unless acoustic echo cancellation (AEC) is performed via hardware or software.
\ No newline at end of file
+While the model was trained to be robust to background noise and reverberation, it will still perform the best when the audio is relatively clean and free of overly loud background noise. In particular, the presence of audio playback of music/speech from the same device that is capturing the microphone stream may result in significantly higher false-reject rates unless acoustic echo cancellation (AEC) is performed via hardware or software.