Update app.py

app.py

@@ -119,20 +119,31 @@ def generate_cmx_edl(edl_title, segments, source_name, fps=25, start_offset=0.0)
     reel_id = source_name.replace(" ", "_")[:8] # EDL reel IDs are traditionally short max 8 chars
     
     for i, seg in enumerate(segments, 1):
-        src_start_val = float(seg.get('src_start', seg.get('start', 0.0)))
-        src_end_val = float(seg.get('src_end', seg.get('end', 0.0)))
+        is_vo = seg.get('type') == 'vo' or 'text' in seg
         
-        # Add offset to source time
-        src_in = format_timecode(src_start_val + start_offset, fps)
-        src_out = format_timecode(src_end_val + start_offset, fps)
+        if is_vo:
+            vo_text = seg.get('text', 'VO')
+            duration = float(seg.get('duration', max(1.0, len(vo_text.split()) / 2.5)))
+            src_in = format_timecode(0, fps)
+            src_out = format_timecode(duration, fps)
+            current_source = "GEN_VO"
+            note_text = f"VO SCRIPT: {vo_text}"
+        else:
+            src_start_val = float(seg.get('src_start', seg.get('start', 0.0)))
+            src_end_val = float(seg.get('src_end', seg.get('end', 0.0)))
+            src_in = format_timecode(src_start_val + start_offset, fps)
+            src_out = format_timecode(src_end_val + start_offset, fps)
+            duration = src_end_val - src_start_val
+            current_source = reel_id
+            note_text = seg.get('note', 'Clip')
         
-        duration = src_end_val - src_start_val
         rec_in = format_timecode(rec_start, fps)
         rec_out = format_timecode(rec_start + duration, fps)
         
-        edl_lines.append(f"{i:03}  {reel_id:8} V     C        {src_in} {src_out} {rec_in} {rec_out}")
-        edl_lines.append(f"* FROM CLIP NAME: {source_name}")
-        edl_lines.append(f"* {seg.get('note', 'Clip')}\n")
+        edl_lines.append(f"{i:03}  {current_source:8} V     C        {src_in} {src_out} {rec_in} {rec_out}")
+        if not is_vo:
+            edl_lines.append(f"* FROM CLIP NAME: {source_name}")
+        edl_lines.append(f"* {note_text}\n")
         
         gap = float(seg.get('gap', 0.0))
         rec_start += duration + gap
@@ -177,53 +188,90 @@ def generate_xml(sequence_name, segments, source_name, fps=25, start_offset=0.0)
     lines.append('\t\t\t<track>')
 
     file_defined = False
+    vo_file_defined = False
     timeline_head_frames = 0
+    
     for i, seg in enumerate(segments, 1):
-        src_start_val = float(seg.get('src_start', seg.get('start', 0.0)))
-        src_end_val = float(seg.get('src_end', seg.get('end', 0.0)))
+        is_vo = seg.get('type') == 'vo' or 'text' in seg
         
-        # IN and OUT refer to the absolute zero-based frame of the media file. 
-        src_in_frames = seconds_to_frames(src_start_val, fps)
-        src_out_frames = seconds_to_frames(src_end_val, fps)
-        duration_frames = src_out_frames - src_in_frames
+        if is_vo:
+            vo_text = seg.get('text', 'VO')
+            duration_sec = float(seg.get('duration', max(1.0, len(vo_text.split()) / 2.5)))
+            duration_frames = seconds_to_frames(duration_sec, fps)
+            src_in_frames = 0
+            src_out_frames = duration_frames
+            master_id_to_use = "masterfile-vo"
+            source_name_to_use = "Generated_VO_Placeholder"
+            clip_note = f"VO SCRIPT: {vo_text}"
+            clip_name_disp = (clip_note[:75] + '..') if len(clip_note) > 75 else clip_note
+        else:
+            src_start_val = float(seg.get('src_start', seg.get('start', 0.0)))
+            src_end_val = float(seg.get('src_end', seg.get('end', 0.0)))
+            src_in_frames = seconds_to_frames(src_start_val, fps)
+            src_out_frames = seconds_to_frames(src_end_val, fps)
+            duration_frames = src_out_frames - src_in_frames
+            master_id_to_use = master_id
+            source_name_to_use = source_name
+            raw_note = seg.get('note', 'Junior Editor Selection')
+            clip_note = raw_note
+            clip_name_disp = source_name
         
         tl_start = timeline_head_frames
         tl_end = tl_start + duration_frames
-        
-        raw_note = seg.get('note', 'Junior Editor Selection')
-        clip_note = (raw_note[:75] + '..') if len(raw_note) > 75 else raw_note
 
         lines.append(f'\t\t\t\t<clipitem id="clipitem-v-{i}">')
-        lines.append(f'\t\t\t\t\t<name>{source_name}</name>')
+        lines.append(f'\t\t\t\t\t<name>{clip_name_disp}</name>')
         lines.append(f'\t\t\t\t\t<duration>8640000</duration>')
         lines.append(f'\t\t\t\t\t<rate><timebase>{timebase}</timebase></rate>')
         lines.append(f'\t\t\t\t\t<start>{tl_start}</start>')
         lines.append(f'\t\t\t\t\t<end>{tl_end}</end>')
         lines.append(f'\t\t\t\t\t<in>{src_in_frames}</in>')
         lines.append(f'\t\t\t\t\t<out>{src_out_frames}</out>')
-        lines.append(f'\t\t\t\t\t<masterclipid>{master_id}</masterclipid>')
+        lines.append(f'\t\t\t\t\t<masterclipid>{master_id_to_use}</masterclipid>')
         
-        # --- ROBUST FILE DEFINITION ---
-        if not file_defined:
-            lines.append(f'\t\t\t\t\t<file id="{master_id}">')
-            lines.append(f'\t\t\t\t\t\t<name>{source_name}</name>')
-            lines.append(f'\t\t\t\t\t\t<pathurl>{path_url}</pathurl>') 
-            lines.append(f'\t\t\t\t\t\t<rate><timebase>{timebase}</timebase></rate>')
-            lines.append(f'\t\t\t\t\t\t<duration>8640000</duration>') 
-            lines.append(f'\t\t\t\t\t\t<timecode>')
-            lines.append(f'\t\t\t\t\t\t\t<rate><timebase>{timebase}</timebase></rate>')
-            lines.append(f'\t\t\t\t\t\t\t<string>{start_tc_string}</string>') 
-            lines.append(f'\t\t\t\t\t\t\t<frame>{start_offset_frames}</frame>')
-            lines.append(f'\t\t\t\t\t\t\t<displayformat>NDF</displayformat>')
-            lines.append(f'\t\t\t\t\t\t</timecode>')
-            lines.append('\t\t\t\t\t\t<media>')
-            lines.append('\t\t\t\t\t\t\t<video><samplecharacteristics><width>1920</width><height>1080</height></samplecharacteristics></video>')
-            lines.append('\t\t\t\t\t\t\t<audio><samplecharacteristics><depth>16</depth><samplerate>48000</samplerate></samplecharacteristics><channelcount>2</channelcount></audio>')
-            lines.append('\t\t\t\t\t\t</media>')
-            lines.append('\t\t\t\t\t</file>')
-            file_defined = True
+        # --- ROBUST FILE DEFINITIONS ---
+        if is_vo:
+            if not vo_file_defined:
+                lines.append(f'\t\t\t\t\t<file id="{master_id_to_use}">')
+                lines.append(f'\t\t\t\t\t\t<name>{source_name_to_use}</name>')
+                lines.append(f'\t\t\t\t\t\t<pathurl>file://localhost/{source_name_to_use}.wav</pathurl>') 
+                lines.append(f'\t\t\t\t\t\t<rate><timebase>{timebase}</timebase></rate>')
+                lines.append(f'\t\t\t\t\t\t<duration>8640000</duration>') 
+                lines.append(f'\t\t\t\t\t\t<timecode>')
+                lines.append(f'\t\t\t\t\t\t\t<rate><timebase>{timebase}</timebase></rate>')
+                lines.append(f'\t\t\t\t\t\t\t<string>00:00:00:00</string>') 
+                lines.append(f'\t\t\t\t\t\t\t<frame>0</frame>')
+                lines.append(f'\t\t\t\t\t\t\t<displayformat>NDF</displayformat>')
+                lines.append(f'\t\t\t\t\t\t</timecode>')
+                lines.append('\t\t\t\t\t\t<media>')
+                lines.append('\t\t\t\t\t\t\t<video><samplecharacteristics><width>1920</width><height>1080</height></samplecharacteristics></video>')
+                lines.append('\t\t\t\t\t\t\t<audio><samplecharacteristics><depth>16</depth><samplerate>48000</samplerate></samplecharacteristics><channelcount>2</channelcount></audio>')
+                lines.append('\t\t\t\t\t\t</media>')
+                lines.append('\t\t\t\t\t</file>')
+                vo_file_defined = True
+            else:
+                lines.append(f'\t\t\t\t\t<file id="{master_id_to_use}"/>')
         else:
-            lines.append(f'\t\t\t\t\t<file id="{master_id}"/>')
+            if not file_defined:
+                lines.append(f'\t\t\t\t\t<file id="{master_id_to_use}">')
+                lines.append(f'\t\t\t\t\t\t<name>{source_name_to_use}</name>')
+                lines.append(f'\t\t\t\t\t\t<pathurl>{path_url}</pathurl>') 
+                lines.append(f'\t\t\t\t\t\t<rate><timebase>{timebase}</timebase></rate>')
+                lines.append(f'\t\t\t\t\t\t<duration>8640000</duration>') 
+                lines.append(f'\t\t\t\t\t\t<timecode>')
+                lines.append(f'\t\t\t\t\t\t\t<rate><timebase>{timebase}</timebase></rate>')
+                lines.append(f'\t\t\t\t\t\t\t<string>{start_tc_string}</string>') 
+                lines.append(f'\t\t\t\t\t\t\t<frame>{start_offset_frames}</frame>')
+                lines.append(f'\t\t\t\t\t\t\t<displayformat>NDF</displayformat>')
+                lines.append(f'\t\t\t\t\t\t</timecode>')
+                lines.append('\t\t\t\t\t\t<media>')
+                lines.append('\t\t\t\t\t\t\t<video><samplecharacteristics><width>1920</width><height>1080</height></samplecharacteristics></video>')
+                lines.append('\t\t\t\t\t\t\t<audio><samplecharacteristics><depth>16</depth><samplerate>48000</samplerate></samplecharacteristics><channelcount>2</channelcount></audio>')
+                lines.append('\t\t\t\t\t\t</media>')
+                lines.append('\t\t\t\t\t</file>')
+                file_defined = True
+            else:
+                lines.append(f'\t\t\t\t\t<file id="{master_id_to_use}"/>')
             
         lines.append('\t\t\t\t\t<marker>')
         lines.append(f'\t\t\t\t\t\t<name>{clip_note}</name>')
@@ -245,25 +293,38 @@ def generate_xml(sequence_name, segments, source_name, fps=25, start_offset=0.0)
 
     timeline_head_frames = 0
     for i, seg in enumerate(segments, 1):
-        src_start_val = float(seg.get('src_start', seg.get('start', 0.0)))
-        src_end_val = float(seg.get('src_end', seg.get('end', 0.0)))
+        is_vo = seg.get('type') == 'vo' or 'text' in seg
         
-        src_in_frames = seconds_to_frames(src_start_val, fps)
-        src_out_frames = seconds_to_frames(src_end_val, fps)
-        duration_frames = src_out_frames - src_in_frames
+        if is_vo:
+            vo_text = seg.get('text', 'VO')
+            duration_sec = float(seg.get('duration', max(1.0, len(vo_text.split()) / 2.5)))
+            duration_frames = seconds_to_frames(duration_sec, fps)
+            src_in_frames = 0
+            src_out_frames = duration_frames
+            master_id_to_use = "masterfile-vo"
+            source_name_to_use = "Generated_VO_Placeholder"
+        else:
+            src_start_val = float(seg.get('src_start', seg.get('start', 0.0)))
+            src_end_val = float(seg.get('src_end', seg.get('end', 0.0)))
+            src_in_frames = seconds_to_frames(src_start_val, fps)
+            src_out_frames = seconds_to_frames(src_end_val, fps)
+            duration_frames = src_out_frames - src_in_frames
+            master_id_to_use = master_id
+            source_name_to_use = source_name
+            
         tl_start = timeline_head_frames
         tl_end = tl_start + duration_frames
 
         lines.append(f'\t\t\t\t<clipitem id="clipitem-a-{i}">')
-        lines.append(f'\t\t\t\t\t<name>{source_name}</name>')
-        lines.append(f'\t\t\t\t\t<masterclipid>{master_id}</masterclipid>')
+        lines.append(f'\t\t\t\t\t<name>{source_name_to_use}</name>')
+        lines.append(f'\t\t\t\t\t<masterclipid>{master_id_to_use}</masterclipid>')
         lines.append(f'\t\t\t\t\t<duration>8640000</duration>')
         lines.append(f'\t\t\t\t\t<rate><timebase>{timebase}</timebase></rate>')
         lines.append(f'\t\t\t\t\t<start>{tl_start}</start>')
         lines.append(f'\t\t\t\t\t<end>{tl_end}</end>')
         lines.append(f'\t\t\t\t\t<in>{src_in_frames}</in>')
         lines.append(f'\t\t\t\t\t<out>{src_out_frames}</out>')
-        lines.append(f'\t\t\t\t\t<file id="{master_id}"/>')
+        lines.append(f'\t\t\t\t\t<file id="{master_id_to_use}"/>')
         lines.append('\t\t\t\t\t<sourcetrack><mediatype>audio</mediatype><trackindex>1</trackindex></sourcetrack>')
         lines.append('\t\t\t\t</clipitem>')
         timeline_head_frames = tl_end + seconds_to_frames(float(seg.get('gap', 0.0)), fps)
@@ -276,21 +337,40 @@ def generate_xml(sequence_name, segments, source_name, fps=25, start_offset=0.0)
     
     return "\n".join(lines)
 
-def generate_transcript_txt(sequence_name, segments, fps=25, start_offset=0.0):
-    """Generates a human-readable text document of the edit decisions and transcription."""
+def generate_transcript_txt(sequence_name, segments, fps=25, start_offset=0.0, transcript_data=None):
+    """Generates a human-readable text document of the edit decisions and exact transcription."""
     lines = [f"Sequence Transcript: {sequence_name}", "=" * 50, ""]
     
     for i, seg in enumerate(segments, 1):
-        src_start_val = float(seg.get('src_start', seg.get('start', 0.0)))
-        src_end_val = float(seg.get('src_end', seg.get('end', 0.0)))
-        
-        src_in = format_timecode(src_start_val + start_offset, fps)
-        src_out = format_timecode(src_end_val + start_offset, fps)
+        is_vo = seg.get('type') == 'vo' or 'text' in seg
         
-        note = seg.get('note', 'No note provided.')
-        
-        lines.append(f"Clip {i:02}  |  IN: {src_in}  -->  OUT: {src_out}")
-        lines.append(f"Text: {note}")
+        if is_vo:
+            vo_text = seg.get('text', 'VO')
+            duration = float(seg.get('duration', max(1.0, len(vo_text.split()) / 2.5)))
+            lines.append(f"Clip {i:02}  |  [GENERATED VOICE-OVER]  |  Est. Duration: {duration:.1f}s")
+            lines.append(f"Script: \"{vo_text}\"")
+        else:
+            src_start_val = float(seg.get('src_start', seg.get('start', 0.0)))
+            src_end_val = float(seg.get('src_end', seg.get('end', 0.0)))
+            src_in = format_timecode(src_start_val + start_offset, fps)
+            src_out = format_timecode(src_end_val + start_offset, fps)
+            note = seg.get('note', 'No note provided.')
+            
+            # Cross-reference the exact times with the original Whisper transcript to get the exact words
+            actual_text = ""
+            if transcript_data:
+                clip_words = []
+                for t_seg in transcript_data:
+                    # Check if the transcript segment overlaps with our clip's time window
+                    if max(src_start_val, t_seg['start']) < min(src_end_val, t_seg['end']):
+                        clip_words.append(t_seg['text'])
+                actual_text = " ".join(clip_words).strip()
+
+            lines.append(f"Clip {i:02}  |  IN: {src_in}  -->  OUT: {src_out}")
+            lines.append(f"AI Summary: {note}")
+            if actual_text:
+                lines.append(f"Exact Audio Text: \"{actual_text}\"")
+            
         lines.append("") # Blank line to separate clips visually
         
     return "\n".join(lines)
@@ -303,15 +383,20 @@ def call_gemini_for_edl(transcript_data, story_prompt, api_key):
         
     system_prompt = (
         "You are an expert Documentary Senior Editor. Use the provided transcript JSON "
-        "(which includes Speaker IDs and word-level timestamps) to create a condensed story. "
-        "Output ONLY a valid JSON array of segments with 'src_start', 'src_end', 'note', and optionally 'gap'. "
-        "CRITICAL RULES:\n"
-        "1. IGNORE ALL INTERVIEWER COMMENTS: Do not include any speech or segments where the interviewer is speaking.\n"
-        "2. REMOVE FLUFF: Delete 'um', 'ah', repeats, and irrelevant filler.\n"
-        "3. NARRATIVE FLOW: Focus on the subject's high-energy responses and narrative hooks.\n"
-        "4. TIMESTAMP INTEGRITY: Use only the exact word-level start and end times from the data.\n"
-        "5. PACING: Group related clips together. Between distinct ideas, add a 'gap': 1.0 (float seconds) "
-        "to the segment preceding the break."
+        "(which includes Speaker IDs and word-level timestamps) to create a condensed story.\n\n"
+        "NEW CAPABILITY: You can now act as a Writer/Director and generate Voice-Over (VO) to bridge gaps, introduce topics, or summarize. "
+        "CRITICAL: ONLY generate 'vo' segments if the user explicitly requests voice over suggestions or narration in their creative brief. Otherwise, ONLY use 'clip' segments.\n\n"
+        "Output ONLY a valid JSON array of segments. Every segment MUST be either a 'clip' or a 'vo'.\n\n"
+        "For 'clip' segments (extracting from the subject):\n"
+        "{\"type\": \"clip\", \"src_start\": 12.5, \"src_end\": 25.0, \"note\": \"Subject talks about X\", \"gap\": 0.0}\n"
+        "- IGNORE ALL INTERVIEWER COMMENTS.\n"
+        "- REMOVE FLUFF: Delete 'um', 'ah', repeats, and irrelevant filler.\n"
+        "- TIMESTAMP INTEGRITY: Use only the exact word-level start and end times from the data.\n\n"
+        "For 'vo' segments (generating new voice-over):\n"
+        "{\"type\": \"vo\", \"text\": \"The journey began years ago...\", \"duration\": 3.5, \"gap\": 0.0}\n"
+        "- DURATION: Estimate duration accurately based on reading speed (approx 2.5 words per second).\n"
+        "- Keep VO concise and in the style of a documentary narrator.\n\n"
+        "PACING: Group related clips. Add a 'gap' (in seconds) between distinct ideas."
     )
     
     prompt_text = f"Creative Brief: {story_prompt}\n\nTranscript Data:\n{json.dumps(transcript_data)}"
@@ -402,6 +487,7 @@ st.markdown("""
 * Upload your file here (video or audio).
 * Set your timeline FPS, Source Timecode, and transcription quality in the sidebar.
 * Junior Editor will transcribe and separate speakers. You can then instruct it to find engaging bits or construct a narrative.
+* **Note: If you want me to write voice over suggestions, please request it.**
 * It will create an EDL or XML to import back into your editing software (Resolve, Premiere).
 * **For Multicam Workflows:** Use the **XML** option in the sidebar and enter the **exact name** of your Multicam Sequence.
 * **Resolve Users:** Uncheck "Automatically import source clips into media pool" during import.
@@ -459,6 +545,8 @@ if uploaded_file:
     if "last_processed_file" not in st.session_state or st.session_state.last_processed_file != uploaded_file.name:
         if "transcript" in st.session_state:
             del st.session_state.transcript
+        if "edit_generated" in st.session_state:
+            del st.session_state.edit_generated
         st.session_state.last_processed_file = uploaded_file.name
 
     # --- Auto-Process Logic ---
@@ -541,7 +629,7 @@ if uploaded_file:
                 st.markdown(f"**{seg['speaker']}:** {seg['text']}")
         
         st.subheader("Your Instruction")
-        brief = st.text_area("What should the Junior Editor do?", placeholder="e.g. Find the most engaging bits and put them together from Speaker 1.")
+        brief = st.text_area("What should the Junior Editor do?", placeholder="e.g. Find the most engaging bits. If you want me to write voice over suggestions, please request it.")
         
         if st.button("Generate Edit"):
             if not ACTIVE_GEMINI_KEY:
@@ -567,22 +655,33 @@ if uploaded_file:
                             final_output = generate_xml(final_source_name, edl_segments, final_source_name, fps, offset_sec)
                             ext = "xml"
                             
-                        # Generate the companion human-readable transcript
-                        transcript_txt = generate_transcript_txt(final_source_name, edl_segments, fps, offset_sec)
-                            
-                        st.subheader("Ready for Import")
-                        st.success(f"✅ Edit generated successfully using **{used_model}** (via {used_endpoint}).")
+                        # Generate the companion human-readable transcript, passing the raw transcript data for exact word matching
+                        transcript_txt = generate_transcript_txt(final_source_name, edl_segments, fps, offset_sec, st.session_state.transcript)
                         
-                        # Display downloads in a neat row
-                        col1, col2 = st.columns(2)
-                        with col1:
-                            st.download_button(f"Download .{ext.upper()} Sequence", data=final_output, file_name=f"{safe_filename}.{ext}")
-                        with col2:
-                            st.download_button(f"Download Reference Transcript (.TXT)", data=transcript_txt, file_name=f"{safe_filename}-Transcript.txt")
+                        # Save to session state so buttons don't disappear on click
+                        st.session_state.edit_generated = True
+                        st.session_state.final_output = final_output
+                        st.session_state.ext = ext
+                        st.session_state.transcript_txt = transcript_txt
+                        st.session_state.safe_filename = safe_filename
+                        st.session_state.used_model = used_model
+                        st.session_state.used_endpoint = used_endpoint
                         
-                        with st.expander(f"Show {ext.upper()} Code", expanded=False):
-                            st.code(final_output, language="xml" if ext == "xml" else "text")
-                        
-                        with st.expander("Show Transcript Reference", expanded=False):
-                            st.text(transcript_txt)
+        # Render the download buttons outside the Generate block using session state
+        if st.session_state.get("edit_generated"):
+            st.subheader("Ready for Import")
+            st.success(f"✅ Edit generated successfully using **{st.session_state.used_model}** (via {st.session_state.used_endpoint}).")
+            
+            # Display downloads in a neat row
+            col1, col2 = st.columns(2)
+            with col1:
+                st.download_button(f"Download .{st.session_state.ext.upper()} Sequence", data=st.session_state.final_output, file_name=f"{st.session_state.safe_filename}.{st.session_state.ext}")
+            with col2:
+                st.download_button(f"Download Reference Transcript (.TXT)", data=st.session_state.transcript_txt, file_name=f"{st.session_state.safe_filename}-Transcript.txt")
+            
+            with st.expander(f"Show {st.session_state.ext.upper()} Code", expanded=False):
+                st.code(st.session_state.final_output, language="xml" if st.session_state.ext == "xml" else "text")
+            
+            with st.expander("Show Transcript Reference", expanded=False):
+                st.text(st.session_state.transcript_txt)