0.1708.2 - chunking buffer increased

FlyingFathead · Dec 20, 2024 · fc86e37 · fc86e37
1 parent cd48d16
commit fc86e37
Show file tree

Hide file tree

Showing 3 changed files with 14 additions and 3 deletions.
diff --git a/README.md b/README.md
@@ -196,6 +196,8 @@ After launching your bot successfully, you can interact with it via Telegram (se
 - `/language` - set the model's transcription language (`auto` =  autodetect); if you know the language spoken in the audio, setting the transcription language manually with this command may improve both transcription speed and accuracy.
 
 ## Changes
+- v0.1708.2 - Added buffer for chunking
+   - Changed the chunk sizes from `4096` to `4000` to avoid edge case 
 - v0.1708.1 - Small bug fixes in the output
    - Note that running the program within `firejail` using Nvidia driver v.560.xx or newer requires i.e.:
    ```

diff --git a/src/main.py b/src/main.py
@@ -3,7 +3,7 @@
 # openai-whisper transcriber-bot for Telegram
 
 # version of this program
-version_number = "0.1708.1"
+version_number = "0.1708.2"
 
 # ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 # https://github.com/FlyingFathead/whisper-transcriber-telegram-bot/

diff --git a/src/transcription_handler.py b/src/transcription_handler.py
@@ -383,7 +383,16 @@ async def transcribe_audio(bot, update, audio_path, output_dir, youtube_url, vid
     log_gpu_utilization()  # Log GPU utilization before starting transcription
 
     logger.info(f"Using device: {device} for transcription")
-    transcription_command = ["whisper", audio_path, "--model", model, "--output_dir", output_dir, "--device", device]
+
+    # transcription_command = ["whisper", audio_path, "--model", model, "--output_dir", output_dir, "--device", device]
+
+    transcription_command = [
+        "whisper", audio_path, 
+        "--model", model, 
+        "--output_dir", output_dir, 
+        "--device", device
+    ]
+
     if language and language != "auto":
         logger.info(f"Starting transcription with model '{model}' and language '{language}' for: {audio_path}")
         transcription_command.extend(["--language", language])
@@ -599,7 +608,7 @@ async def process_url_message(message_text, bot, update, model, language):
             # Here is where we add the transcription_note
             transcription_note = "📝🔊 <i>(transcribed audio)</i>\n\n"
             note_length = len(transcription_note)
-            max_message_length = 4096 - note_length  # Adjust max length to account for transcription note
+            max_message_length = 4000 - note_length  # Adjust max length to account for transcription note
 
             if transcription_settings['send_as_messages'] and 'txt' in transcription_paths:
                 try: