NickVerri commited on
Commit
ae15678
·
verified ·
1 Parent(s): 8f2c23a

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +46 -25
app.py CHANGED
@@ -7,8 +7,9 @@ import requests
7
  from datetime import timedelta
8
 
9
  # --- Configuration ---
10
- # This key should be set in your Hugging Face Space Secrets
11
- API_KEY = os.environ.get("GEMINI_API_KEY", "")
 
12
 
13
  def format_timecode(seconds, fps=25):
14
  td = timedelta(seconds=seconds)
@@ -34,11 +35,11 @@ def generate_cmx_edl(edl_title, segments, source_name, fps=25):
34
  rec_start += duration
35
  return "\n".join(edl_lines)
36
 
37
- def call_gemini_for_edl(transcript_data, story_prompt):
38
- if not API_KEY:
39
- st.error("Gemini API Key not found in Environment Variables. Please set GEMINI_API_KEY in Space Secrets.")
40
  return None
41
- url = f"https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash-preview-09-2025:generateContent?key={API_KEY}"
42
  system_prompt = (
43
  "You are an expert Documentary Senior Editor. Use the transcript JSON to create a condensed story. "
44
  "Output ONLY a valid JSON array of segments with 'src_start', 'src_end', and 'note'. "
@@ -61,42 +62,62 @@ def call_gemini_for_edl(transcript_data, story_prompt):
61
  st.set_page_config(page_title="DocAI Editor", layout="wide")
62
  st.title("Documentary AI: Senior Editor Pipeline")
63
 
 
 
 
 
 
 
 
 
 
 
 
 
64
  uploaded_file = st.file_uploader("Upload Video/Audio", type=["mp4", "m4a", "wav", "mp3", "mov"])
65
- fps = st.sidebar.number_input("Timeline FPS", value=25)
66
 
67
  if uploaded_file:
68
  if "transcript" not in st.session_state:
69
  if st.button("Step 1: Transcribe"):
70
- with st.spinner("Processing... Stripping audio and transcribing (this may take a few minutes):"):
71
  # Save uploaded file
72
  with open("temp_input", "wb") as f:
73
  f.write(uploaded_file.getbuffer())
74
 
75
- # Extract audio using FFmpeg
76
  subprocess.run([
77
- "ffmpeg", "-i", "temp_input", "-vn", "-acodec", "pcm_s16le",
78
- "-ar", "44100", "-ac", "2", "audio.wav", "-y"
 
 
 
 
 
79
  ])
80
 
81
  # Transcribe with Whisper
82
  model = whisper.load_model("base")
83
- result = model.transcribe("audio.wav", word_timestamps=True)
84
  st.session_state.transcript = result['segments']
85
  st.success("Transcription Complete.")
86
 
87
  if "transcript" in st.session_state:
88
  st.divider()
89
- brief = st.text_area("Creative Brief", placeholder="e.g., Create a 2-minute edit focusing on the 'uncorking the brain' story and the importance of public education.")
 
90
  if st.button("Step 2: Create EDL"):
91
- with st.spinner("Senior Editor analyzing transcript..."):
92
- segments = call_gemini_for_edl(st.session_state.transcript, brief)
93
- if segments:
94
- edl_content = generate_cmx_edl("AI_Edit", segments, uploaded_file.name, fps)
95
- st.subheader("EDL Preview")
96
- st.text_area("Raw CMX 3600 Output", value=edl_content, height=250)
97
- st.download_button(
98
- label="Download EDL for Resolve",
99
- data=edl_content,
100
- file_name=f"{os.path.splitext(uploaded_file.name)[0]}_AI_Edit.edl",
101
- mime="text/plain"
102
- )
 
 
 
 
7
  from datetime import timedelta
8
 
9
  # --- Configuration ---
10
+ # Check for key in environment secrets first (Hugging Face / Docker)
11
+ # If not found, we will use the sidebar input
12
+ ENV_API_KEY = os.environ.get("GEMINI_API_KEY", "")
13
 
14
  def format_timecode(seconds, fps=25):
15
  td = timedelta(seconds=seconds)
 
35
  rec_start += duration
36
  return "\n".join(edl_lines)
37
 
38
+ def call_gemini_for_edl(transcript_data, story_prompt, api_key):
39
+ if not api_key:
40
+ st.error("Gemini API Key is missing. Please provide it in the sidebar.")
41
  return None
42
+ url = f"https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash-preview-09-2025:generateContent?key={api_key}"
43
  system_prompt = (
44
  "You are an expert Documentary Senior Editor. Use the transcript JSON to create a condensed story. "
45
  "Output ONLY a valid JSON array of segments with 'src_start', 'src_end', and 'note'. "
 
62
  st.set_page_config(page_title="DocAI Editor", layout="wide")
63
  st.title("Documentary AI: Senior Editor Pipeline")
64
 
65
+ # Sidebar for Key and Settings
66
+ with st.sidebar:
67
+ st.header("Settings")
68
+ # Priority: User input > Environment Secret
69
+ user_key = st.text_input("Gemini API Key", type="password", help="Paste your Gemini API key here.")
70
+ active_api_key = user_key if user_key else ENV_API_KEY
71
+
72
+ if not active_api_key:
73
+ st.warning("⚠️ No API Key found. Enter it above to enable Step 2.")
74
+
75
+ fps = st.number_input("Timeline FPS", value=25)
76
+
77
  uploaded_file = st.file_uploader("Upload Video/Audio", type=["mp4", "m4a", "wav", "mp3", "mov"])
 
78
 
79
  if uploaded_file:
80
  if "transcript" not in st.session_state:
81
  if st.button("Step 1: Transcribe"):
82
+ with st.spinner("Processing... Extracting optimized audio and transcribing:"):
83
  # Save uploaded file
84
  with open("temp_input", "wb") as f:
85
  f.write(uploaded_file.getbuffer())
86
 
87
+ # Optimized FFmpeg for voice: m4a, 64kbps, 16kHz, mono
88
  subprocess.run([
89
+ "ffmpeg", "-i", "temp_input",
90
+ "-vn", # No video
91
+ "-acodec", "aac", # AAC codec for m4a
92
+ "-ab", "64k", # 64kbps bitrate
93
+ "-ar", "16000", # 16kHz sample rate
94
+ "-ac", "1", # Mono (single channel)
95
+ "audio.m4a", "-y"
96
  ])
97
 
98
  # Transcribe with Whisper
99
  model = whisper.load_model("base")
100
+ result = model.transcribe("audio.m4a", word_timestamps=True)
101
  st.session_state.transcript = result['segments']
102
  st.success("Transcription Complete.")
103
 
104
  if "transcript" in st.session_state:
105
  st.divider()
106
+ brief = st.text_area("Creative Brief", placeholder="e.g., Create a 2-minute edit focusing on the 'uncorking the brain' story.")
107
+
108
  if st.button("Step 2: Create EDL"):
109
+ if not active_api_key:
110
+ st.error("Please provide a Gemini API Key in the sidebar to proceed.")
111
+ else:
112
+ with st.spinner("Senior Editor analyzing transcript..."):
113
+ segments = call_gemini_for_edl(st.session_state.transcript, brief, active_api_key)
114
+ if segments:
115
+ edl_content = generate_cmx_edl("AI_Edit", segments, uploaded_file.name, fps)
116
+ st.subheader("EDL Preview")
117
+ st.text_area("Raw CMX 3600 Output", value=edl_content, height=250)
118
+ st.download_button(
119
+ label="Download EDL for Resolve",
120
+ data=edl_content,
121
+ file_name=f"{os.path.splitext(uploaded_file.name)[0]}_AI_Edit.edl",
122
+ mime="text/plain"
123
+ )