Update app.py
Browse files
app.py
CHANGED
@@ -88,6 +88,46 @@ def detect_landmark(image, detector, predictor):
|
|
88 |
coords[i] = (shape.part(i).x, shape.part(i).y)
|
89 |
return coords
|
90 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
91 |
def preprocess_video(input_video_path):
|
92 |
if torch.cuda.is_available():
|
93 |
detector = dlib.cnn_face_detection_model_v1(face_detector_path)
|
@@ -189,8 +229,8 @@ with demo:
|
|
189 |
detect_landmark_btn.click(preprocess_video, [video_in], [
|
190 |
video_out])
|
191 |
predict_btn = gr.Button("Predict")
|
192 |
-
predict_btn.click(predict, [video_out], [
|
193 |
-
|
194 |
with gr.Row():
|
195 |
# video_lip = gr.Video(label="Audio Visual Video", mirror_webcam=False)
|
196 |
text_output.render()
|
|
|
88 |
coords[i] = (shape.part(i).x, shape.part(i).y)
|
89 |
return coords
|
90 |
|
91 |
+
def predict_and_save(process_video):
|
92 |
+
num_frames = int(cv2.VideoCapture(process_video).get(cv2.CAP_PROP_FRAME_COUNT))
|
93 |
+
|
94 |
+
tsv_cont = ["/\n", f"test-0\t{process_video}\t{None}\t{num_frames}\t{int(16_000*num_frames/25)}\n"]
|
95 |
+
label_cont = ["DUMMY\n"]
|
96 |
+
with open(f"{data_dir}/test.tsv", "w") as fo:
|
97 |
+
fo.write("".join(tsv_cont))
|
98 |
+
with open(f"{data_dir}/test.wrd", "w") as fo:
|
99 |
+
fo.write("".join(label_cont))
|
100 |
+
task.load_dataset(gen_subset, task_cfg=saved_cfg.task)
|
101 |
+
|
102 |
+
def decode_fn(x):
|
103 |
+
dictionary = task.target_dictionary
|
104 |
+
symbols_ignore = generator.symbols_to_strip_from_output
|
105 |
+
symbols_ignore.add(dictionary.pad())
|
106 |
+
return task.datasets[gen_subset].label_processors[0].decode(x, symbols_ignore)
|
107 |
+
|
108 |
+
itr = task.get_batch_iterator(dataset=task.dataset(gen_subset)).next_epoch_itr(shuffle=False)
|
109 |
+
sample = next(itr)
|
110 |
+
if torch.cuda.is_available():
|
111 |
+
sample = utils.move_to_cuda(sample)
|
112 |
+
hypos = task.inference_step(generator, models, sample)
|
113 |
+
ref = decode_fn(sample['target'][0].int().cpu())
|
114 |
+
hypo = hypos[0][0]['tokens'].int().cpu()
|
115 |
+
hypo = decode_fn(hypo)
|
116 |
+
|
117 |
+
# Collect timestamps and texts
|
118 |
+
transcript = []
|
119 |
+
for i, (start, end) in enumerate(sample['net_input']['video_lengths'], 1):
|
120 |
+
start_time = float(start) / 16_000
|
121 |
+
end_time = float(end) / 16_000
|
122 |
+
text = hypo[i].strip()
|
123 |
+
transcript.append({"timestamp": [start_time, end_time], "text": text})
|
124 |
+
|
125 |
+
# Save transcript to a JSON file
|
126 |
+
with open('speech_transcript.json', 'w') as outfile:
|
127 |
+
json.dump(transcript, outfile, indent=4)
|
128 |
+
|
129 |
+
return hypo
|
130 |
+
|
131 |
def preprocess_video(input_video_path):
|
132 |
if torch.cuda.is_available():
|
133 |
detector = dlib.cnn_face_detection_model_v1(face_detector_path)
|
|
|
229 |
detect_landmark_btn.click(preprocess_video, [video_in], [
|
230 |
video_out])
|
231 |
predict_btn = gr.Button("Predict")
|
232 |
+
#predict_btn.click(predict, [video_out], [text_output])
|
233 |
+
predict_btn.click(predict_and_save, [video_out], [text_output])
|
234 |
with gr.Row():
|
235 |
# video_lip = gr.Video(label="Audio Visual Video", mirror_webcam=False)
|
236 |
text_output.render()
|