Update app.py
Browse files
app.py
CHANGED
|
@@ -5,7 +5,7 @@ import json
|
|
| 5 |
import os
|
| 6 |
import requests
|
| 7 |
import torch
|
| 8 |
-
import numpy
|
| 9 |
from datetime import timedelta
|
| 10 |
from pyannote.audio import Pipeline
|
| 11 |
from huggingface_hub import hf_hub_download
|
|
@@ -158,7 +158,27 @@ if uploaded_file:
|
|
| 158 |
if torch.cuda.is_available():
|
| 159 |
pipeline.to(torch.device("cuda"))
|
| 160 |
|
| 161 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 162 |
except Exception as e:
|
| 163 |
st.error(f"Diarization Error: {e}")
|
| 164 |
st.stop()
|
|
@@ -172,11 +192,17 @@ if uploaded_file:
|
|
| 172 |
for segment in result['segments']:
|
| 173 |
mid_time = (segment['start'] + segment['end']) / 2
|
| 174 |
speaker = "Unknown"
|
| 175 |
-
for turn, _, speaker_id in diarization.itertracks(yield_label=True):
|
| 176 |
-
if turn.start <= mid_time <= turn.end:
|
| 177 |
-
speaker = speaker_id
|
| 178 |
-
break
|
| 179 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 180 |
final_segments.append({
|
| 181 |
"speaker": speaker,
|
| 182 |
"text": segment['text'],
|
|
@@ -194,4 +220,19 @@ if uploaded_file:
|
|
| 194 |
# Diarization Preview to help the user identify roles
|
| 195 |
with st.expander("Transcript Preview (Diarized)"):
|
| 196 |
for seg in st.session_state.transcript[:20]: # Show first 20 segments
|
| 197 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 5 |
import os
|
| 6 |
import requests
|
| 7 |
import torch
|
| 8 |
+
import numpy
|
| 9 |
from datetime import timedelta
|
| 10 |
from pyannote.audio import Pipeline
|
| 11 |
from huggingface_hub import hf_hub_download
|
|
|
|
| 158 |
if torch.cuda.is_available():
|
| 159 |
pipeline.to(torch.device("cuda"))
|
| 160 |
|
| 161 |
+
diarization_output = pipeline("audio_optimized.m4a")
|
| 162 |
+
|
| 163 |
+
# Fix for 'DiarizeOutput' object has no attribute 'itertracks'
|
| 164 |
+
# Handle potential tuple output or direct object
|
| 165 |
+
if isinstance(diarization_output, tuple):
|
| 166 |
+
diarization = diarization_output[0]
|
| 167 |
+
else:
|
| 168 |
+
diarization = diarization_output
|
| 169 |
+
|
| 170 |
+
# Verify we have the annotation object by checking for itertracks
|
| 171 |
+
if not hasattr(diarization, "itertracks"):
|
| 172 |
+
# If it's a Hook/Wrapper object (common in some versions), try getting the annotation
|
| 173 |
+
if hasattr(diarization_output, "annotation"):
|
| 174 |
+
diarization = diarization_output.annotation
|
| 175 |
+
else:
|
| 176 |
+
# Last resort: if it behaves like a dict, try getting 'annotation' key
|
| 177 |
+
try:
|
| 178 |
+
diarization = diarization_output['annotation']
|
| 179 |
+
except:
|
| 180 |
+
pass
|
| 181 |
+
|
| 182 |
except Exception as e:
|
| 183 |
st.error(f"Diarization Error: {e}")
|
| 184 |
st.stop()
|
|
|
|
| 192 |
for segment in result['segments']:
|
| 193 |
mid_time = (segment['start'] + segment['end']) / 2
|
| 194 |
speaker = "Unknown"
|
|
|
|
|
|
|
|
|
|
|
|
|
| 195 |
|
| 196 |
+
# Safe iteration over tracks
|
| 197 |
+
try:
|
| 198 |
+
for turn, _, speaker_id in diarization.itertracks(yield_label=True):
|
| 199 |
+
if turn.start <= mid_time <= turn.end:
|
| 200 |
+
speaker = speaker_id
|
| 201 |
+
break
|
| 202 |
+
except AttributeError:
|
| 203 |
+
# Fallback if diarization object is still malformed
|
| 204 |
+
speaker = "Error-No-Speaker"
|
| 205 |
+
|
| 206 |
final_segments.append({
|
| 207 |
"speaker": speaker,
|
| 208 |
"text": segment['text'],
|
|
|
|
| 220 |
# Diarization Preview to help the user identify roles
|
| 221 |
with st.expander("Transcript Preview (Diarized)"):
|
| 222 |
for seg in st.session_state.transcript[:20]: # Show first 20 segments
|
| 223 |
+
# FIXED: Ensure this line is a complete string with no breaks
|
| 224 |
+
st.write(f"**{seg['speaker']}:** {seg['text']}")
|
| 225 |
+
|
| 226 |
+
brief = st.text_area("Creative Brief", placeholder="e.g. Focus on the yeast story, remove the interviewer.")
|
| 227 |
+
|
| 228 |
+
if st.button("Step 2: Create EDL"):
|
| 229 |
+
if not ACTIVE_GEMINI_KEY:
|
| 230 |
+
st.error("Gemini API Key required. Please set it in Secrets or app.py.")
|
| 231 |
+
else:
|
| 232 |
+
with st.spinner("Analyzing..."):
|
| 233 |
+
edl_segments = call_gemini_for_edl(st.session_state.transcript, brief, ACTIVE_GEMINI_KEY)
|
| 234 |
+
if edl_segments:
|
| 235 |
+
final_edl = generate_cmx_edl("AI_Senior_Editor_Cut", edl_segments, uploaded_file.name, fps)
|
| 236 |
+
st.subheader("EDL Preview")
|
| 237 |
+
st.code(final_edl, language="text")
|
| 238 |
+
st.download_button("Download EDL", data=final_edl, file_name="edit.edl")
|