Update app.py
Browse files
app.py
CHANGED
|
@@ -137,17 +137,16 @@ interface = gr.Interface(
|
|
| 137 |
The final result is chosen by majority vote, using confidence scores to break ties.
|
| 138 |
The output shows the final label, its confidence, and the predictions for each chunk.\n"
|
| 139 |
"\nHow the Model Makes Predictions
|
|
|
|
|
|
|
| 140 |
|
| 141 |
-
|
| 142 |
-
The final result is based on:
|
| 143 |
|
| 144 |
-
|
| 145 |
|
| 146 |
-
|
| 147 |
|
| 148 |
-
|
| 149 |
-
|
| 150 |
-
The output shows the final label, its confidence, and the per-chunk predictions.\n"
|
| 151 |
),
|
| 152 |
)
|
| 153 |
|
|
|
|
| 137 |
The final result is chosen by majority vote, using confidence scores to break ties.
|
| 138 |
The output shows the final label, its confidence, and the predictions for each chunk.\n"
|
| 139 |
"\nHow the Model Makes Predictions
|
| 140 |
+
The audio is split into 5-second chunks and each chunk is turned into a Mel-spectrogram. A CNN predicts a label and confidence score for every chunk.
|
| 141 |
+
The final result is based on:
|
| 142 |
|
| 143 |
+
Majority vote β the class that appears most often across chunks.
|
|
|
|
| 144 |
|
| 145 |
+
Tie-breaker β if two or more classes appear the same number of times, the model selects the one with the highest total confidence across its chunks.
|
| 146 |
|
| 147 |
+
Final confidence β the average confidence of all chunks predicted as the final class.
|
| 148 |
|
| 149 |
+
The output shows the final label, its confidence, and the per-chunk predictions.\n"
|
|
|
|
|
|
|
| 150 |
),
|
| 151 |
)
|
| 152 |
|