Create app.py
Browse files
app.py
ADDED
|
@@ -0,0 +1,59 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import streamlit as st
|
| 2 |
+
st.title("🎥 Video Subtitle Generator with Chroma DB and Cosine Similarity")
|
| 3 |
+
|
| 4 |
+
# Upload video
|
| 5 |
+
uploaded_file = st.file_uploader("Upload a video", type=["mp4", "avi", "mov", "mkv"])
|
| 6 |
+
|
| 7 |
+
if uploaded_file:
|
| 8 |
+
with tempfile.NamedTemporaryFile(delete=False, suffix=".mp4") as temp_video:
|
| 9 |
+
temp_video.write(uploaded_file.getbuffer())
|
| 10 |
+
video_path = temp_video.name
|
| 11 |
+
|
| 12 |
+
audio_path = "temp_audio.wav"
|
| 13 |
+
|
| 14 |
+
# Extract audio
|
| 15 |
+
st.info("Extracting audio...")
|
| 16 |
+
extract_audio(video_path, audio_path)
|
| 17 |
+
|
| 18 |
+
# Transcribe audio
|
| 19 |
+
st.info("Transcribing audio...")
|
| 20 |
+
transcribed_text = transcribe_audio(audio_path)
|
| 21 |
+
st.text_area("Transcribed Text", transcribed_text, height=150)
|
| 22 |
+
|
| 23 |
+
# Initialize Chroma DB client
|
| 24 |
+
chroma_client = chromadb.Client()
|
| 25 |
+
|
| 26 |
+
# Load subtitle database into Chroma DB
|
| 27 |
+
subtitle_db_path = "database.csv"
|
| 28 |
+
collection, df = load_subtitle_db_chroma(subtitle_db_path, chroma_client)
|
| 29 |
+
|
| 30 |
+
# Find matching subtitles with Chroma DB
|
| 31 |
+
st.info("Finding matching subtitles...")
|
| 32 |
+
matching_subtitles, subtitle_embeddings = find_chroma_subtitles(transcribed_text, collection)
|
| 33 |
+
|
| 34 |
+
# Generate query embedding for cosine similarity
|
| 35 |
+
query_embedding = generate_embedding(transcribed_text)
|
| 36 |
+
|
| 37 |
+
# Compute cosine similarity
|
| 38 |
+
cosine_similarities = compute_cosine_similarity(query_embedding, subtitle_embeddings)
|
| 39 |
+
|
| 40 |
+
# Sort by cosine similarity
|
| 41 |
+
for i, sub in enumerate(matching_subtitles):
|
| 42 |
+
sub['cosine_similarity'] = cosine_similarities[i]
|
| 43 |
+
|
| 44 |
+
# Sort by similarity score
|
| 45 |
+
matching_subtitles = sorted(matching_subtitles, key=lambda x: x['cosine_similarity'], reverse=True)
|
| 46 |
+
|
| 47 |
+
# Display video
|
| 48 |
+
st.video(video_path)
|
| 49 |
+
|
| 50 |
+
# Display matching subtitles with similarity scores
|
| 51 |
+
st.subheader("📜 Matching Subtitles (Chroma DB + Cosine Similarity)")
|
| 52 |
+
for sub in matching_subtitles:
|
| 53 |
+
st.write(f"**Subtitle:** {sub['text']}")
|
| 54 |
+
st.write(f"**Cosine Similarity:** {sub['cosine_similarity']:.4f}")
|
| 55 |
+
st.write("---")
|
| 56 |
+
|
| 57 |
+
# Cleanup
|
| 58 |
+
os.remove(video_path)
|
| 59 |
+
os.remove(audio_path)
|