.env_example

@@ -1,6 +1,7 @@
-OPENAI_API_KEY = 
+INPUT_VIDEO_URL=
-OPENAI_BASE_URL = 
+OPENAI_API_KEY=
-OPENAI_RESPONSES_PROMPT = 
+OPENAI_BASE_URL=
-SEGMENT_DURATION = 
+OPENAI_TRANSCRIPTION_MODEL=
-TMP_AUDIO_PATH = 
+OPENAI_CHAT_SYSTEM_PROMPT=
-TMP_VIDEO_PATH = 
+OPENAI_CHAT_MODEL=
+OPENAI_CHAT_N=

.gitignore

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

app.py

@@ -1,15 +1,26 @@
 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."
+DEFAULT_PROMPT = "The user will provided a video transcription for which you are to generate a blog post in Markdown format summarizing the video contents.  Please only output the blog post content."
+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(args):
+def main():
-    openai_client = OpenAI()
+    openai_client = OpenAI(
+            base_url = OPENAI_BASE_URL,
+            api_key = OPENAI_API_KEY
+            )
     return summarize_transcription(
             openai_client,
             transcribe_audio(
@@ -21,39 +32,51 @@ def main(args):
             )
 
 def get_video_from_url():
-    video_file_url = os.getenv('INPUT_VIDEO_URL')
+    filename = VIDEO_URL.split('/')[-1]
-    video_file_path = os.getenv('TMP_VIDEO_PATH' , '/tmp/video_summary_bot_tmp_video.mp4')
+    with open(f"{OUTPUT_PATH}/{filename}", 'wb') as f:
-    request = requests.get(video_file_url)
+        for chunk in requests.get(VIDEO_URL).iter_content(chunk_size=255):
-    with open(video_file_path, 'wb') as f:
-        for chunk in requests.get(video_file_url).iter_content(chunk_size=255):
             if chunk:
                 f.write(chunk)
+    return filename
 
-def get_audio_from_video():
+def get_audio_from_video(video_filename):
-    tmp_audio_path = os.getenv('TMP_AUDIO_PATH', '/tmp/video_summary_bot_tmp_audio.wav')
+    VideoFileClip(f"{OUTPUT_PATH}/{video_filename}").audio.write_audiofile(f"{OUTPUT_PATH}/{video_filename}.wav")
-    video_file_path = os.getenv('TMP_VIDEO_PATH')
+    audio = AudioSegment.from_wav(f"{OUTPUT_PATH}/{video_filename}.wav")
-    VideoFileClip(video_file_path).audio.write_audiofile(tmp_audio_path)
+    segments = []
-    return AudioSegment.from_wav(tmp_audio_path)
+    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):
+def transcribe_audio(openai_client, audio_segments):
-    segment_duration = int(os.getenv('SEGMENT_DURATION', 30000)),
-    transcription_model = os.getenv('OPENAI_TRANSCRIPTION_MODEL', 'whisper-1')
     return ' '.join([
         openai_client.audio.transcriptions.create(
-            model=transcription_model,
+            model=OPENAI_TRANSCRIPTION_MODEL,
-            file=each
+            file=open(each, 'rb')
-            ).text for each in [audio[i:i + segment_duration] for i in range(0, len(audio), segment_duration)]
+            ).text for each in audio_segments
         ])
 
 def summarize_transcription(openai_client, transcription):
-    prompt = os.getenv('OPENAI_RESPONSES_PROMPT', DEFAULT_PROMPT)
+    return openai_client.chat.completions.create(
-    responses_model = os.getenv('OPENAI_RESPONSES_MODEL', 'whisper-1')
+            model=OPENAI_CHAT_MODEL, 
-    return client.responses.create(
+            n=OPENAI_CHAT_N,
-            model=responses_model,
+            messages = [
-            instructions=prompt,
+                {"role": "developer", "content": OPENAI_CHAT_SYSTEM_PROMPT},
-            input=transcription
+                {"role": "user", "content": transcription}
-            )
+                ]
+            ).choices
+
+def setup():
+    from dotenv import load_dotenv
+    load_dotenv()
+
+def cleanup():
+    os.rmdir(OUTPUT_PATH)
 
 if __name__ == '__main__':
-    load_dotenv()
+    setup()
-    main(parser.parse_args())
+    for each in main():
+        print("========")
+        print(each.message.content)