edits

generate-docker-compose.py

@@ -17,7 +17,7 @@ for channel in channels:
         "environment": [
             f"CHANNEL_NAME={channel['name']}",
             f"CHANNEL_LANGUAGE={channel['language']}",
-            "TIMEDELTA_DAYS=1",
+            "TIMEDELTA_DAYS=2",
             "TIMEDELTA_DAYS_EXACT=true",
             "CLIP_CREATE_FROM_CHAT=false",
             "TWITCH_CLIENT_ID=a0fuj6tm5ct79clvim9816orphqkov",

main.py

@@ -18,24 +18,24 @@ TIMEDELTA_DAYS_EXACT = os.environ.get("TIMEDELTA_DAYS_EXACT", "false").lower() i
 CLIP_CREATE_FROM_CHAT = os.environ.get("CLIP_CREATE_FROM_CHAT", "false").lower() in ("true", "1", "yes")
 CHANNEL_LANGUAGE = os.environ.get("CHANNEL_LANGUAGE", "cs")
 SEARCH_KEYWORDS = [
-  "madmonq",
-  "madmonge",
-  "madmong",
-  "medmong",
-  "medmonk",
-  "madmonk",
-  "mad monk",
-  "mad monq",
-  "mad-monq",
-  "mad-monk",
-  "madmonck",
-  "madmunk",
-  "madmon",
-  "madmonke",
-  "madmonque",
-  "matmonk",
-  "matt monk"
-  "mat monk"
+    "madmonq",
+    "madmonge",
+    "madmong",
+    "medmong",
+    "medmonk",
+    "madmonk",
+    "mad monk",
+    "mad monq",
+    "mad-monq",
+    "mad-monk",
+    "madmonck",
+    "madmunk",
+    "madmon",
+    "madmonke",
+    "madmonque",
+    "matmonk",
+    "matt monk",
+    "mat monk"
 ]
 MODEL_NAME = "turbo"  # Whisper model
 
@@ -144,11 +144,34 @@ def transcribe_audio(audio_file, model_name):
     result = model.transcribe(audio_file, language=CHANNEL_LANGUAGE)
     return result
 
-def transcribe_audio_fast(audio_file, model_name):
+def transcribe_audio_fast(audio_file, model_name, language, vod_id):
+
+    transcript_path = os.path.join(base_dirs["transcripts"], f"transcript_{vod_id}.json")
+    if os.path.exists(transcript_path):
+        print(f"faster_whisper -- Loading existing transcription for VOD {vod_id} from {transcript_path}")
+        with open(transcript_path, "r", encoding="utf-8") as f:
+            segments_data = json.load(f)
+        return segments_data
+
+    # Initialize the model and transcribe (passing language if provided)
     model_fast = WhisperModel("large-v3-turbo", device="auto", compute_type="int8", download_root="/app/models")
-    segments, info = model_fast.transcribe(audio_file)
+    segments, info = model_fast.transcribe(audio_file, language=language)
     print("faster_whisper -- Detected language '%s' with probability %f" % (info.language, info.language_probability))
-    return segments
+
+    # Build a list of dictionaries for the segments.
+    segments_data = []
+    for seg in segments:
+        segments_data.append({
+            "start": seg.start,
+            "end": seg.end,
+            "text": seg.text
+        })
+
+    with open(transcript_path, "w", encoding="utf-8") as f:
+        json.dump(segments_data, f, ensure_ascii=False, indent=4)
+    print(f"faster_whisper -- Saved transcription to {transcript_path}")
+
+    return segments_data
 
 def search_transcription(result, keywords):
     matches = []
@@ -161,7 +184,6 @@ def search_transcription(result, keywords):
                     break  # Stop checking further keywords for this segment
     return matches
 
-
 def scrape_chat_log(vod_id, output_filename):
     """
     Uses TwitchDownloaderCLI to download the chat log for a given VOD.
@@ -215,28 +237,20 @@ def create_clip_from_vod(video_file, match_start, vod):
 def find_comments_by_keywords(chat_log, keywords):
     """
     Searches the chat log for any comments containing one of the given keywords.
-    The chat log can be either:
-      - a raw list of comment objects, or
-      - an object with a "comments" key containing the list.
-    Each comment is expected to have:
-      - a "message" key with the comment text (as a string)
-      - an "offset" key (or fallback to "content_offset_seconds") for the timestamp.
     Returns a list of matching comment objects.
     """
     matching_comments = []
-    # If the chat log is wrapped in an object, extract the list.
     if isinstance(chat_log, dict) and "comments" in chat_log:
         chat_log = chat_log["comments"]
 
     for comment in chat_log:
         if not isinstance(comment, dict):
             continue
-        # Get the message text; TwitchDownloaderCLI outputs it as a string in "message"
         message_text = comment['message']['body'].lower()
         for keyword in keywords:
             if keyword.lower() in message_text:
                 matching_comments.append(comment)
-                break  # No need to check further keywords for this comment.
+                break
     return matching_comments
 
 def create_clip_from_comment_timestamp(video_file, comment_timestamp, vod):
@@ -268,10 +282,10 @@ def create_clip_from_comment_timestamp(video_file, comment_timestamp, vod):
 # ---------------------------
 # Main Processing Pipeline
 # ---------------------------
-def handle_matches_fast(vod, video_filename, result):
+def handle_matches_fast(vod, video_filename, segments_data):
     matches_fast = []
-    for segment in result:
-        segment_text = segment.text.lower()
+    for segment in segments_data:
+        segment_text = segment["text"].lower()
         for keyword in SEARCH_KEYWORDS:
             if keyword.lower() in segment_text:
                 matches_fast.append(segment)
@@ -280,14 +294,13 @@ def handle_matches_fast(vod, video_filename, result):
     if matches_fast:
         print(f"faster_whisper -- Found {len(matches_fast)} mention(s) of {SEARCH_KEYWORDS} in VOD {vod['id']}:")
         for match in matches_fast:
-            start = match.start  # faster-whisper segment attribute
-            text = match.text
+            start = match["start"]
+            text = match["text"]
             print(f" - At {start:.2f}s: {text}")
             create_clip_from_vod(video_filename, start, vod)
     else:
         print("faster_whisper -- No mentions of keywords.")
 
-
 def handle_matches(vod, video_filename, result):
     matches = search_transcription(result, SEARCH_KEYWORDS)
     if matches:
@@ -300,8 +313,6 @@ def handle_matches(vod, video_filename, result):
     else:
         print(f"No mentions of {SEARCH_KEYWORDS} found in VOD {vod['id']}.")
 
-
-
 def main():
     print("Obtaining access token...")
     token = get_access_token()
@@ -330,30 +341,16 @@ def main():
         download_vod(vod_url, video_filename)
         extract_audio(video_filename, audio_filename)
 
-        # # Check if transcript already exists; if yes, load it, otherwise transcribe and save.
-        # if os.path.exists(transcript_filename):
-        #     print(f"{transcript_filename} already exists. Skipping transcription.")
-        #     with open(transcript_filename, "r", encoding="utf-8") as f:
-        #         result = json.load(f)
-        # else:
-        #     print("Transcribing audio. This may take some time...")
-        #     result = transcribe_audio(audio_filename, MODEL_NAME)
-        #     with open(transcript_filename, "w", encoding="utf-8") as f:
-        #         json.dump(result, f, ensure_ascii=False, indent=4)
-        #     print(f"Transcript saved to {transcript_filename}")
-
         print("Transcribing audio. This may take some time...")
-        result = transcribe_audio_fast(audio_filename, MODEL_NAME)
+        # Pass language and vod_id so that the transcript is saved and reused if available.
+        segments_data = transcribe_audio_fast(audio_filename, MODEL_NAME, language=CHANNEL_LANGUAGE, vod_id=vod_id)
 
         if CLIP_CREATE_FROM_CHAT:
             scrape_chat_log(vod_id, chat_log_filename)
 
-        # Search transcript for keywords
-        # handle_matches(vod_id, video_filename, result)
-        handle_matches_fast(vod_id, video_filename, result)
+        handle_matches_fast(vod, video_filename, segments_data)
 
         if CLIP_CREATE_FROM_CHAT:
-            # Load chat log from file
             try:
                 with open(chat_log_filename, "r", encoding="utf-8") as f:
                     chat_log = json.load(f)