Update app.py

app.py

@@ -115,7 +115,323 @@ def generate_cmx_edl(edl_title, segments, source_name, fps=25):
 def generate_xml(sequence_name, segments, source_name, fps=25):
     """Constructs a Final Cut Pro 7 XML with Video AND Audio Track 1."""
     
-    xml_output = [
-        '<?xml version="1.0" encoding="UTF-8"?>',
-        '<!DOCTYPE xmeml>',
-        '<x
+    # Use append mode to avoid copy-paste line break errors
+    lines = []
+    lines.append('<?xml version="1.0" encoding="UTF-8"?>')
+    lines.append('<!DOCTYPE xmeml>')
+    lines.append('<xmeml version="4">')
+    lines.append('<sequence>')
+    lines.append(f'\t<name>{sequence_name}</name>')
+    lines.append('\t<rate>')
+    lines.append(f'\t\t<timebase>{fps}</timebase>')
+    lines.append('\t</rate>')
+    lines.append('\t<media>')
+    
+    # --- VIDEO TRACK ---
+    lines.append('\t\t<video>')
+    lines.append('\t\t\t<format>')
+    lines.append('\t\t\t\t<samplecharacteristics>')
+    lines.append(f'\t\t\t\t\t<rate><timebase>{fps}</timebase></rate>')
+    lines.append('\t\t\t\t\t<width>1920</width>')
+    lines.append('\t\t\t\t\t<height>1080</height>')
+    lines.append('\t\t\t\t\t<pixelaspectratio>square</pixelaspectratio>')
+    lines.append('\t\t\t\t</samplecharacteristics>')
+    lines.append('\t\t\t</format>')
+    lines.append('\t\t\t<track>')
+
+    # --- VIDEO LOOP ---
+    timeline_head_frames = 0
+    for i, seg in enumerate(segments, 1):
+        src_in_frames = seconds_to_frames(seg['src_start'], fps)
+        src_out_frames = seconds_to_frames(seg['src_end'], fps)
+        duration_frames = src_out_frames - src_in_frames
+        
+        tl_start = timeline_head_frames
+        tl_end = tl_start + duration_frames
+        clip_note = seg.get('note', 'Junior Editor Selection')
+
+        lines.append(f'\t\t\t\t<clipitem id="clipitem-v-{i}">')
+        lines.append(f'\t\t\t\t\t<name>{clip_note}</name>')
+        lines.append(f'\t\t\t\t\t<duration>{duration_frames}</duration>')
+        lines.append(f'\t\t\t\t\t<rate><timebase>{fps}</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>')
+        
+        # File Reference
+        lines.append(f'\t\t\t\t\t<file id="multicam_file">')
+        lines.append(f'\t\t\t\t\t\t<name>{source_name}</name>')
+        lines.append(f'\t\t\t\t\t\t<pathurl>file://localhost/placeholder/{source_name}</pathurl>')
+        lines.append(f'\t\t\t\t\t\t<rate><timebase>{fps}</timebase></rate>')
+        lines.append(f'\t\t\t\t\t\t<timecode><string>00:00:00:00</string></timecode>')
+        lines.append(f'\t\t\t\t\t</file>')
+        lines.append(f'\t\t\t\t</clipitem>')
+        
+        gap_seconds = seg.get('gap', 0.0)
+        gap_frames = seconds_to_frames(gap_seconds, fps)
+        timeline_head_frames = tl_end + gap_frames
+
+    lines.append('\t\t\t</track>')
+    lines.append('\t\t</video>')
+    
+    # --- AUDIO TRACK ---
+    lines.append('\t\t<audio>')
+    lines.append('\t\t\t<numOutputChannels>2</numOutputChannels>')
+    lines.append('\t\t\t<format>')
+    lines.append('\t\t\t\t<samplecharacteristics>')
+    lines.append('\t\t\t\t\t<depth>16</depth>')
+    lines.append('\t\t\t\t\t<samplerate>48000</samplerate>')
+    lines.append('\t\t\t\t</samplecharacteristics>')
+    lines.append('\t\t\t</format>')
+    lines.append('\t\t\t<track>') # Audio Track 1
+    
+    # --- AUDIO LOOP (Identical Timing to Video) ---
+    timeline_head_frames = 0
+    for i, seg in enumerate(segments, 1):
+        src_in_frames = seconds_to_frames(seg['src_start'], fps)
+        src_out_frames = seconds_to_frames(seg['src_end'], fps)
+        duration_frames = src_out_frames - src_in_frames
+        
+        tl_start = timeline_head_frames
+        tl_end = tl_start + duration_frames
+        clip_note = seg.get('note', 'Junior Editor Selection')
+
+        lines.append(f'\t\t\t\t<clipitem id="clipitem-a-{i}">')
+        lines.append(f'\t\t\t\t\t<name>{clip_note}</name>')
+        lines.append(f'\t\t\t\t\t<duration>{duration_frames}</duration>')
+        lines.append(f'\t\t\t\t\t<rate><timebase>{fps}</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>')
+        
+        # File Reference (Same ID as video)
+        lines.append(f'\t\t\t\t\t<file id="multicam_file"/>')
+        
+        # SOURCE TRACK MAPPING (Use Source Track 1)
+        lines.append('\t\t\t\t\t<sourcetrack>')
+        lines.append('\t\t\t\t\t\t<mediatype>audio</mediatype>')
+        lines.append('\t\t\t\t\t\t<trackindex>1</trackindex>')
+        lines.append('\t\t\t\t\t</sourcetrack>')
+        
+        lines.append(f'\t\t\t\t</clipitem>')
+        
+        gap_seconds = seg.get('gap', 0.0)
+        gap_frames = seconds_to_frames(gap_seconds, fps)
+        timeline_head_frames = tl_end + gap_frames
+
+    lines.append('\t\t\t</track>')
+    lines.append('\t\t</audio>')
+    lines.append('\t</media>')
+    lines.append('</sequence>')
+    lines.append('</xmeml>')
+    
+    return "\n".join(lines)
+
+def call_gemini_for_edl(transcript_data, story_prompt, api_key):
+    if not api_key:
+        st.error("Gemini API Key is missing.")
+        return None
+        
+    url = f"https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash-preview-09-2025:generateContent?key={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."
+    )
+    
+    prompt_text = f"Creative Brief: {story_prompt}\n\nTranscript Data:\n{json.dumps(transcript_data)}"
+    
+    payload = {
+        "contents": [{"parts": [{"text": prompt_text}]}],
+        "systemInstruction": {"parts": [{"text": system_prompt}]},
+        "generationConfig": {"responseMimeType": "application/json"}
+    }
+    
+    try:
+        res = requests.post(url, json=payload)
+        res.raise_for_status()
+        result_json = res.json()
+        return json.loads(result_json['candidates'][0]['content']['parts'][0]['text'])
+    except Exception as e:
+        st.error(f"Senior Editor AI Error: {e}")
+        return None
+
+# --- Streamlit UI ---
+st.set_page_config(page_title="Junior Editor", layout="wide")
+st.title("Junior Editor")
+
+st.markdown("""
+**Instructions**
+* Upload your file here (video or audio).
+* Set your timeline FPS and transcription quality.
+* Junior Editor will transcribe and separate speakers. You can then instruct it to find engaging bits or construct a narrative.
+* 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.
+""")
+
+st.divider()
+
+with st.sidebar:
+    st.header("Project Settings")
+    fps = st.number_input("Timeline FPS", value=25)
+    
+    st.header("Export Settings")
+    export_format = st.radio("Output Format", ["EDL", "XML (Multicam)"], index=0)
+    
+    input_label = "EDL Reel Name"
+    input_help = "Leave empty to use the uploaded file name."
+    
+    if export_format == "XML (Multicam)":
+        input_label = "Multicam Sequence Name"
+        input_help = "EXACT name of your Multicam Clip in Resolve."
+    
+    st.info("💡 **Conform Helper**")
+    custom_reel_name = st.text_input(
+        input_label, 
+        placeholder="e.g. Interview_Day1_Multi", 
+        help=input_help
+    )
+
+    st.header("Model Settings")
+    model_size = st.selectbox("Whisper Model", ["large-v2", "medium", "base"], index=0)
+    
+    language_map = {
+        "Auto-Detect": None,
+        "English": "en",
+        "Spanish": "es",
+        "French": "fr",
+        "German": "de",
+        "Italian": "it",
+        "Portuguese": "pt"
+    }
+    selected_lang_label = st.selectbox("Audio Language", list(language_map.keys()), index=1)
+    target_language = language_map[selected_lang_label]
+    
+    num_speakers = st.number_input("Speakers (0=Auto)", min_value=0, value=0)
+    
+    st.divider()
+    if ACTIVE_HF_TOKEN == "PASTE_YOUR_HF_TOKEN_HERE":
+        st.warning("⚠️ HF_TOKEN not set in Secrets!")
+    else:
+        st.success("✅ HF_TOKEN Loaded")
+
+uploaded_file = st.file_uploader("Upload Video/Audio Clip", type=["mp4", "m4a", "wav", "mp3", "mov"])
+
+if uploaded_file:
+    # --- Auto-Reset Logic ---
+    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
+        st.session_state.last_processed_file = uploaded_file.name
+
+    # --- Auto-Process Logic ---
+    if "transcript" not in st.session_state:
+        if not ACTIVE_HF_TOKEN or "PASTE_YOUR_HF_TOKEN" in ACTIVE_HF_TOKEN:
+            st.error("Please provide a valid Hugging Face Token in the Sidebar/Secrets.")
+        else:
+            progress_container = st.container()
+            with progress_container:
+                st.info("🤖 **Junior Editor is processing your file...**")
+                status_text = st.empty()
+                progress_bar = st.progress(0)
+                
+                try:
+                    # Phase 1
+                    status_text.markdown("**Phase 1/4: Extracting Audio...**")
+                    with open("temp_input", "wb") as f:
+                        f.write(uploaded_file.getbuffer())
+                    
+                    subprocess.run(["ffmpeg", "-i", "temp_input", "-vn", "-acodec", "pcm_s16le", "-ar", "16000", "-ac", "1", "temp_audio.wav", "-y"])
+                    progress_bar.progress(25)
+                    
+                    device = "cuda" if torch.cuda.is_available() else "cpu"
+                    if device == "cpu": st.warning("⚠️ No GPU detected.")
+                    
+                    # Phase 2
+                    status_text.markdown(f"**Phase 2/4: Transcribing (Whisper {model_size})... This is the longest step.**")
+                    compute_type = "float16" if device == "cuda" else "int8"
+                    model = whisperx.load_model(model_size, device, compute_type=compute_type)
+                    audio = whisperx.load_audio("temp_audio.wav")
+                    result = model.transcribe(audio, batch_size=16, language=target_language)
+                    del model
+                    gc.collect()
+                    torch.cuda.empty_cache()
+                    progress_bar.progress(50)
+                    
+                    # Phase 3
+                    status_text.markdown("**Phase 3/4: Aligning Text...**")
+                    model_a, metadata = whisperx.load_align_model(language_code=result["language"], device=device)
+                    result = whisperx.align(result["segments"], model_a, metadata, audio, device, return_char_alignments=False)
+                    del model_a
+                    gc.collect()
+                    torch.cuda.empty_cache()
+                    progress_bar.progress(75)
+                    
+                    # Phase 4
+                    status_text.markdown("**Phase 4/4: Identifying Speakers...**")
+                    diarize_model = whisperx.DiarizationPipeline(use_auth_token=ACTIVE_HF_TOKEN, device=device)
+                    diarize_kwargs = {"min_speakers": num_speakers, "max_speakers": num_speakers} if num_speakers > 0 else {}
+                    diarize_segments = diarize_model(audio, **diarize_kwargs)
+                    
+                    # Final Merge
+                    status_text.markdown("**Finalizing...**")
+                    final_result = whisperx.assign_word_speakers(diarize_segments, result)
+                    
+                    processed_segments = []
+                    for segment in final_result["segments"]:
+                        processed_segments.append({
+                            "speaker": segment.get("speaker", "Unknown"),
+                            "text": segment["text"].strip(),
+                            "start": segment["start"],
+                            "end": segment["end"]
+                        })
+                    
+                    st.session_state.transcript = processed_segments
+                    if os.path.exists("temp_input"): os.remove("temp_input")
+                    if os.path.exists("temp_audio.wav"): os.remove("temp_audio.wav")
+                    progress_bar.progress(100)
+                    status_text.success(f"Done! Processed {len(processed_segments)} segments.")
+
+                except Exception as e:
+                    status_text.error(f"Error: {e}")
+                    if os.path.exists("temp_input"): os.remove("temp_input")
+                    st.stop()
+
+    if "transcript" in st.session_state:
+        st.divider()
+        with st.expander("Transcript Preview", expanded=True):
+            for seg in st.session_state.transcript: 
+                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.")
+        
+        if st.button("Generate Edit"):
+            if not ACTIVE_GEMINI_KEY:
+                st.error("Gemini API Key required.")
+            else:
+                with st.spinner("Junior Editor is thinking..."):
+                    final_source_name = custom_reel_name.strip() if custom_reel_name.strip() else uploaded_file.name
+                    
+                    edl_segments = call_gemini_for_edl(st.session_state.transcript, brief, ACTIVE_GEMINI_KEY)
+                    if edl_segments:
+                        if export_format == "EDL":
+                            final_output = generate_cmx_edl("Junior_Editor_Cut", edl_segments, final_source_name, fps)
+                            ext = "edl"
+                        else:
+                            final_output = generate_xml("Junior_Editor_Cut", edl_segments, final_source_name, fps)
+                            ext = "xml"
+                            
+                        st.subheader("Ready for Import")
+                        st.code(final_output, language="xml" if ext == "xml" else "text")
+                        st.download_button(f"Download .{ext.upper()}", data=final_output, file_name=f"junior_editor_cut.{ext}")