add top_n

app.py

@@ -75,7 +75,11 @@ description = "An audio event classifier trained on the AudioSet dataset to pred
 
 demo = gr.Interface(
     predict_uri,
-    inputs=[gr.inputs.Audio(type="filepath"), gr.inputs.Audio(source="microphone", type="filepath")],
+    inputs=[
+        gr.inputs.Audio(type="filepath"),
+        gr.inputs.Audio(source="microphone", type="filepath"),
+        gr.Slider(minimum=7, maximum=21)
+    ],
     outputs=['image', 'image', 'image', 'text', 'text', 'text', 'text'],
     # examples=examples,
     title=title,

test.py

@@ -19,7 +19,7 @@ OUT_SAMPLE_RATE = 16000
 OUT_PCM = 'PCM_16'
 CLASS_MAP_FILE = 'res/yamnet_class_map.csv'
 DEBUG = True
-SNORING_TOP_N = 7
+# SNORING_TOP_N = 21
 SNORING_INDEX = 38
 IN_MODEL_SAMPLES = 15600
 
@@ -68,13 +68,13 @@ def scores_to_index(scores, order):
     return np.argsort(means, axis=0)[order]
 
 
-def predict_waveform(idx, waveform):
+def predict_waveform(idx, waveform, top_n):
     # Download the YAMNet class map (see main YAMNet model docs) to yamnet_class_map.csv
     # See YAMNet TF2 usage sample for class_names_from_csv() definition.
     scores = predict(MODEL_PATH, waveform)
     class_names = class_names_from_csv(CLASS_MAP_FILE)
 
-    top_n = SNORING_TOP_N
+    # top_n = SNORING_TOP_N
     top_n_res = ''
     snoring_score = 0.0
     for n in range(1, top_n):
@@ -98,15 +98,15 @@ def to_float32(data):
     return np.float32(data)
 
 
-def predict_float32(idx, data):
-    return predict_waveform(idx, to_float32(data))
+def predict_float32(idx, data, top_n):
+    return predict_waveform(idx, to_float32(data), top_n)
 
 
 def split_given_size(arr, size):
     return np.split(arr, np.arange(size, len(arr), size))
 
 
-def predict_uri(audio_uri1, audio_uri2):
+def predict_uri(audio_uri1, audio_uri2, top_n):
     result = ''
     if DEBUG: print('audio_uri1:', audio_uri1, 'audio_uri2:', audio_uri2)
 
@@ -129,7 +129,7 @@ def predict_uri(audio_uri1, audio_uri2):
         second_start = idx * predict_seconds
         result += (int_to_min_sec(second_start) + ', ')
         if len(split) == predict_samples:
-            print_result, snoring_score = predict_float32(idx, split)
+            print_result, snoring_score = predict_float32(idx, split, top_n)
             result += print_result
             snoring_scores.append(snoring_score)
 
@@ -147,9 +147,9 @@ def predict_uri(audio_uri1, audio_uri2):
     apnea_sec = second_total - snoring_sec
     apnea_frequency = (apnea_sec / 10) / second_total
     ahi_result = str(
-        'snoring_sec:' + str(snoring_sec) + ', apnea_sec:' + str(apnea_sec) + ', second_total:' + str(second_total)
-        + ', snoring_frequency:' + format_float(snoring_frequency)
-        + ', apnea_frequency:' + format_float(apnea_frequency)
+        '打鼾秒数snoring_sec=' + str(snoring_sec) + ', 暂停秒数apnea_sec=' + str(apnea_sec) + ', 总秒数second_total=' + str(second_total)
+        + ', 打鼾频率snoring_frequency=' + str(snoring_sec) + '/' + str(second_total) + '=' + format_float(snoring_frequency)
+        + ', 暂停频率apnea_frequency=（' + str(apnea_sec) + '/' + str(10) + ')/' + str(second_total) + '=' + format_float(apnea_frequency)
     )
 
     return waveform_line, mfcc_line, mfcc2_line, str(ahi_result), str(snoring_booleans), str(snoring_scores), str(result)