Execute application with Gitea Actions (#4)

Closes Broadcast/video-summary-bot#1
Closes Broadcast/video-summary-bot#3

Reviewed-on: #4

.env_example

@@ -1,5 +1,7 @@
-OPENAI_API_KEY = 
-OPENAI_BASE_URL = 
-OPENAI_RESPONSES_PROMPT = 
-SEGMENT_DURATION = 
-TMP_AUDIO_PATH = 
+INPUT_VIDEO_URL=
+OPENAI_API_KEY=
+OPENAI_BASE_URL=
+OPENAI_TRANSCRIPTION_MODEL=
+OPENAI_CHAT_SYSTEM_PROMPT=
+OPENAI_CHAT_MODEL=
+OPENAI_CHAT_N=

.gitea/workflows/execute.yaml

@@ -0,0 +1,29 @@
+name: execute
+
+on:
+  workflow_dispatch:
+    inputs:
+      video-url:
+        description: "URL for the video to be analyzed"
+        required: true
+
+jobs:
+  Python:
+    runs-on: ubuntu-latest
+    env:
+      OPENAI_API_KEY: "nokey"
+      OPENAI_BASE_URL: "http://192.168.1.168/v1"
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.10"
+
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip
+          pip install -r requirements.txt
+
+      - name: Run applicaiton
+        run: python app.py

.gitignore

@@ -1,2 +1,4 @@
+__pycache__/*
+tmp/*
 venv/*
 .env

app.py

@@ -1,49 +1,79 @@
-import argparse
 import os
+import requests
 
-from dotenv import load_dotenv
 from moviepy import VideoFileClip
 from openai import OpenAI
 from pydub import AudioSegment
 
-DEFAULT_PROMPT = "You will be provided a video transcription for which you are to generate a blog post in Markdown format summarizing the video contents."
+from prompt import DEFAULT_PROMPT
 
-def main(args):
-    openai_client = OpenAI()
+VIDEO_URL = os.getenv('INPUT_VIDEO_URL', None)
+OUTPUT_PATH = os.getenv('OUTPUT_PATH', 'tmp')
+AUDIO_SEGMENT_DURATION = 30000
+OPENAI_API_KEY = os.getenv('OPENAI_API_KEY', None)
+OPENAI_BASE_URL = os.getenv('OPENAI_BASE_URL', 'https://api.openai.com/v1')
+OPENAI_TRANSCRIPTION_MODEL = os.getenv('OPENAI_TRANSCRIPTION_MODEL', 'whisper-1')
+OPENAI_CHAT_SYSTEM_PROMPT = os.getenv('OPENAI_CHAT_SYSTEM_PROMPT', DEFAULT_PROMPT)
+OPENAI_CHAT_MODEL = os.getenv('OPENAI_CHAT_MODEL', 'whisper-1')
+OPENAI_CHAT_N = int(os.getenv('OPENAI_CHAT_N', '3'))
+
+def main():
+    openai_client = OpenAI(
+            base_url = OPENAI_BASE_URL,
+            api_key = OPENAI_API_KEY
+            )
     return summarize_transcription(
             openai_client,
             transcribe_audio(
                 openai_client,
-                get_audio_from_video(args.video_file_path)
+                get_audio_from_video(
+                    get_video_from_url()
+                    )
                 )
             )
 
-def get_audio_from_video(video_file_path):
-    tmp_audio_path = os.getenv('TMP_AUDIO_PATH', '/tmp/video_summary_bot_tmp_audio.wav')
-    VideoFileClip(video_file_path).audio.write_audiofile(tmp_audio_path)
-    return AudioSegment.from_wav(tmp_audio_path)
+def get_video_from_url():
+    filename = VIDEO_URL.split('/')[-1]
+    with open(f"{OUTPUT_PATH}/{filename}", 'wb') as f:
+        for chunk in requests.get(VIDEO_URL).iter_content(chunk_size=255):
+            if chunk:
+                f.write(chunk)
+    return filename
 
-def transcribe_audio(openai_client, audio):
-    segment_duration = int(os.getenv('SEGMENT_DURATION', 30000)),
-    transcription_model = os.getenv('OPENAI_TRANSCRIPTION_MODEL', 'whisper-1')
+def get_audio_from_video(video_filename):
+    VideoFileClip(f"{OUTPUT_PATH}/{video_filename}").audio.write_audiofile(f"{OUTPUT_PATH}/{video_filename}.wav")
+    audio = AudioSegment.from_wav(f"{OUTPUT_PATH}/{video_filename}.wav")
+    segments = []
+    for i in range(0, len(audio), AUDIO_SEGMENT_DURATION):
+        segment = audio[i:i + AUDIO_SEGMENT_DURATION]
+        path = f"{OUTPUT_PATH}/audio_segment_{i // AUDIO_SEGMENT_DURATION}.wav"
+        segments.append(path)
+        segment.export(path, format='wav')
+    return segments
+
+def transcribe_audio(openai_client, audio_segments):
     return ' '.join([
         openai_client.audio.transcriptions.create(
-            model=transcription_model,
-            file=each
-            ).text for each in [audio[i:i + segment_duration] for i in range(0, len(audio), segment_duration)]
+            model=OPENAI_TRANSCRIPTION_MODEL,
+            file=open(each, 'rb')
+            ).text for each in audio_segments
         ])
 
 def summarize_transcription(openai_client, transcription):
-    prompt = os.getenv('OPENAI_RESPONSES_PROMPT', DEFAULT_PROMPT)
-    responses_model = os.getenv('OPENAI_RESPONSES_MODEL', 'whisper-1')
-    return client.responses.create(
-            model=responses_model,
-            instructions=prompt,
-            input=transcription
-            )
+    return openai_client.completions.create(
+            model=OPENAI_CHAT_MODEL, 
+            prompt=OPENAI_CHAT_SYSTEM_PROMPT.format(transcription)
+            ).choices
+
+def setup():
+    from dotenv import load_dotenv
+    load_dotenv()
+
+def cleanup():
+    os.rmdir(OUTPUT_PATH)
 
 if __name__ == '__main__':
-    load_dotenv()
-    parser = argparse.ArgumentParser(description="Use AI models to summarize videos")
-    parser.add_argument('--video-file-path', type=str, help="Path to the video to be summarized")
-    main(parser.parse_args())
+    setup()
+    for each in main():
+        print("========")
+        print(each.text)

prompt.py

@@ -0,0 +1,19 @@
+DEFAULT_PROMPT = """
+You are a professional blog writer and SEO expert.  You will be given the transcript of a live stream for which you are to generate
+a blog post.
+            
+Instructions:
+    - The blog post title should be SEO optimized.
+    - The blog post should be properly and beautifully formatted using markdown.
+    - Each blog post should have around 5 sections with 3 sub-sections each.
+    - Each sub section should have about 3 paragraphs.
+    - Sub-section headings should be clearly marked.
+    - Ensure that the content flows logically from one section to another, maintaining coherence and readability.
+    - In the final section, provide a forward-looking perspective on the topic and a conclusion.
+    - Make the blog post sound as human and as engaging as possible, add real world examples and make it as informative as possible.
+    - Please ensure proper and standard markdown formatting always.
+
+Transcription: {}
+
+Blog Post: 
+"""

requirements.txt

@@ -1,2 +1,5 @@
 moviepy
+openai
 pydub
+python-dotenv
+requests