fix bug for when transcript length is 1 and combine_transcripts was skipping last segment by using len()-1
Browse files- handler.py +7 -11
handler.py
CHANGED
@@ -49,9 +49,9 @@ class EndpointHandler():
|
|
49 |
video_with_transcript = self.transcribe_video(video_url)
|
50 |
encode_transcript = data.pop("encode_transcript", True)
|
51 |
if encode_transcript:
|
52 |
-
|
53 |
encoded_segments = {
|
54 |
-
"encoded_segments": self.encode_sentences(
|
55 |
}
|
56 |
return {
|
57 |
**video_with_transcript,
|
@@ -112,18 +112,14 @@ class EndpointHandler():
|
|
112 |
all_batches = []
|
113 |
for i in tqdm(range(0, len(transcripts), batch_size)):
|
114 |
# find end position of batch (for when we hit end of data)
|
115 |
-
i_end = min(len(transcripts)
|
116 |
# extract the metadata like text, start/end positions, etc
|
117 |
batch_meta = [{
|
118 |
-
**
|
119 |
-
} for
|
120 |
# extract only text to be encoded by embedding model
|
121 |
batch_text = [
|
122 |
-
row['text'] for row in
|
123 |
-
]
|
124 |
-
# extract IDs to be attached to each embedding and metadata
|
125 |
-
batch_ids = [
|
126 |
-
row['id'] for row in transcripts[i:i_end]
|
127 |
]
|
128 |
# create the embedding vectors
|
129 |
batch_vectors = self.sentence_transformer_model.encode(batch_text).tolist()
|
@@ -152,7 +148,7 @@ class EndpointHandler():
|
|
152 |
video_info = video['video']
|
153 |
transcript_segments = video['transcript']['segments']
|
154 |
for i in tqdm(range(0, len(transcript_segments), stride)):
|
155 |
-
i_end = min(len(transcript_segments)
|
156 |
text = ' '.join(transcript['text']
|
157 |
for transcript in
|
158 |
transcript_segments[i:i_end])
|
|
|
49 |
video_with_transcript = self.transcribe_video(video_url)
|
50 |
encode_transcript = data.pop("encode_transcript", True)
|
51 |
if encode_transcript:
|
52 |
+
encoded_segments = self.combine_transcripts(video_with_transcript)
|
53 |
encoded_segments = {
|
54 |
+
"encoded_segments": self.encode_sentences(encoded_segments)
|
55 |
}
|
56 |
return {
|
57 |
**video_with_transcript,
|
|
|
112 |
all_batches = []
|
113 |
for i in tqdm(range(0, len(transcripts), batch_size)):
|
114 |
# find end position of batch (for when we hit end of data)
|
115 |
+
i_end = min(len(transcripts), i + batch_size)
|
116 |
# extract the metadata like text, start/end positions, etc
|
117 |
batch_meta = [{
|
118 |
+
**row
|
119 |
+
} for row in transcripts[i:i_end]]
|
120 |
# extract only text to be encoded by embedding model
|
121 |
batch_text = [
|
122 |
+
row['text'] for row in batch_meta
|
|
|
|
|
|
|
|
|
123 |
]
|
124 |
# create the embedding vectors
|
125 |
batch_vectors = self.sentence_transformer_model.encode(batch_text).tolist()
|
|
|
148 |
video_info = video['video']
|
149 |
transcript_segments = video['transcript']['segments']
|
150 |
for i in tqdm(range(0, len(transcript_segments), stride)):
|
151 |
+
i_end = min(len(transcript_segments), i + window)
|
152 |
text = ' '.join(transcript['text']
|
153 |
for transcript in
|
154 |
transcript_segments[i:i_end])
|