oceansweep's picture
Upload 155 files
43cd37c verified
# Audio_Files.py
#########################################
# Audio Processing Library
# This library is used to download or load audio files from a local directory.
#
####
#
# Functions:
#
# download_audio_file(url, save_path)
# process_audio(
# process_audio_file(audio_url, audio_file, whisper_model="small.en", api_name=None, api_key=None)
#
#
#########################################
# Imports
import json
import logging
import os
import subprocess
import tempfile
import time
import uuid
from datetime import datetime
from pathlib import Path
#
# External Imports
import requests
import yt_dlp
#
# Local Imports
from App_Function_Libraries.DB.DB_Manager import add_media_with_keywords, \
check_media_and_whisper_model
from App_Function_Libraries.Metrics.metrics_logger import log_counter, log_histogram
from App_Function_Libraries.Summarization.Summarization_General_Lib import perform_summarization
from App_Function_Libraries.Utils.Utils import downloaded_files, \
sanitize_filename, generate_unique_id, temp_files
from App_Function_Libraries.Video_DL_Ingestion_Lib import extract_metadata
from App_Function_Libraries.Audio.Audio_Transcription_Lib import speech_to_text
from App_Function_Libraries.Chunk_Lib import improved_chunking_process
#
#######################################################################################################################
# Function Definitions
#
MAX_FILE_SIZE = 500 * 1024 * 1024
def download_audio_file(url, current_whisper_model="", use_cookies=False, cookies=None):
try:
# Check if media already exists in the database and compare whisper models
should_download, reason = check_media_and_whisper_model(
url=url,
current_whisper_model=current_whisper_model
)
if not should_download:
logging.info(f"Skipping audio download: {reason}")
return None
logging.info(f"Proceeding with audio download: {reason}")
# Set up the request headers
headers = {}
if use_cookies and cookies:
try:
cookie_dict = json.loads(cookies)
headers['Cookie'] = '; '.join([f'{k}={v}' for k, v in cookie_dict.items()])
except json.JSONDecodeError:
logging.warning("Invalid cookie format. Proceeding without cookies.")
# Make the request
response = requests.get(url, headers=headers, stream=True)
# Raise an exception for bad status codes
response.raise_for_status()
# Get the file size
file_size = int(response.headers.get('content-length', 0))
if file_size > 500 * 1024 * 1024: # 500 MB limit
raise ValueError("File size exceeds the 500MB limit.")
# Generate a unique filename
file_name = f"audio_{uuid.uuid4().hex[:8]}.mp3"
save_path = os.path.join('downloads', file_name)
# Ensure the downloads directory exists
os.makedirs('downloads', exist_ok=True)
# Download the file
with open(save_path, 'wb') as f:
for chunk in response.iter_content(chunk_size=8192):
if chunk:
f.write(chunk)
logging.info(f"Audio file downloaded successfully: {save_path}")
return save_path
except requests.RequestException as e:
logging.error(f"Error downloading audio file: {str(e)}")
raise
except ValueError as e:
logging.error(str(e))
raise
except Exception as e:
logging.error(f"Unexpected error downloading audio file: {str(e)}")
raise
def process_audio_files(audio_urls, audio_file, whisper_model, api_name, api_key, use_cookies, cookies, keep_original,
custom_keywords, custom_prompt_input, chunk_method, max_chunk_size, chunk_overlap,
use_adaptive_chunking, use_multi_level_chunking, chunk_language, diarize,
keep_timestamps, custom_title):
start_time = time.time() # Start time for processing
processed_count = 0
failed_count = 0
progress = []
all_transcriptions = []
all_summaries = []
#v2
def format_transcription_with_timestamps(segments):
if keep_timestamps:
formatted_segments = []
for segment in segments:
start = segment.get('Time_Start', 0)
end = segment.get('Time_End', 0)
text = segment.get('Text', '').strip() # Ensure text is stripped of leading/trailing spaces
# Add the formatted timestamp and text to the list, followed by a newline
formatted_segments.append(f"[{start:.2f}-{end:.2f}] {text}")
# Join the segments with a newline to ensure proper formatting
return "\n".join(formatted_segments)
else:
# Join the text without timestamps
return "\n".join([segment.get('Text', '').strip() for segment in segments])
def update_progress(message):
progress.append(message)
return "\n".join(progress)
def cleanup_files():
for file in temp_files:
try:
if os.path.exists(file):
os.remove(file)
update_progress(f"Temporary file {file} removed.")
except Exception as e:
update_progress(f"Failed to remove temporary file {file}: {str(e)}")
def reencode_mp3(mp3_file_path):
try:
reencoded_mp3_path = mp3_file_path.replace(".mp3", "_reencoded.mp3")
subprocess.run([ffmpeg_cmd, '-i', mp3_file_path, '-codec:a', 'libmp3lame', reencoded_mp3_path], check=True)
update_progress(f"Re-encoded {mp3_file_path} to {reencoded_mp3_path}.")
return reencoded_mp3_path
except subprocess.CalledProcessError as e:
update_progress(f"Error re-encoding {mp3_file_path}: {str(e)}")
raise
def convert_mp3_to_wav(mp3_file_path):
try:
wav_file_path = mp3_file_path.replace(".mp3", ".wav")
subprocess.run([ffmpeg_cmd, '-i', mp3_file_path, wav_file_path], check=True)
update_progress(f"Converted {mp3_file_path} to {wav_file_path}.")
return wav_file_path
except subprocess.CalledProcessError as e:
update_progress(f"Error converting {mp3_file_path} to WAV: {str(e)}")
raise
try:
# Check and set the ffmpeg command
global ffmpeg_cmd
if os.name == "nt":
logging.debug("Running on Windows")
ffmpeg_cmd = os.path.join(os.getcwd(), "Bin", "ffmpeg.exe")
else:
ffmpeg_cmd = 'ffmpeg' # Assume 'ffmpeg' is in PATH for non-Windows systems
# Ensure ffmpeg is accessible
if not os.path.exists(ffmpeg_cmd) and os.name == "nt":
raise FileNotFoundError(f"ffmpeg executable not found at path: {ffmpeg_cmd}")
# Define chunk options early to avoid undefined errors
chunk_options = {
'method': chunk_method,
'max_size': max_chunk_size,
'overlap': chunk_overlap,
'adaptive': use_adaptive_chunking,
'multi_level': use_multi_level_chunking,
'language': chunk_language
}
# Process multiple URLs
urls = [url.strip() for url in audio_urls.split('\n') if url.strip()]
for i, url in enumerate(urls):
update_progress(f"Processing URL {i + 1}/{len(urls)}: {url}")
# Download and process audio file
audio_file_path = download_audio_file(url, use_cookies, cookies)
if not os.path.exists(audio_file_path):
update_progress(f"Downloaded file not found: {audio_file_path}")
failed_count += 1
log_counter(
metric_name="audio_files_failed_total",
labels={"whisper_model": whisper_model, "api_name": api_name},
value=1
)
continue
temp_files.append(audio_file_path)
update_progress("Audio file downloaded successfully.")
# Re-encode MP3 to fix potential issues
reencoded_mp3_path = reencode_mp3(audio_file_path)
if not os.path.exists(reencoded_mp3_path):
update_progress(f"Re-encoded file not found: {reencoded_mp3_path}")
failed_count += 1
log_counter(
metric_name="audio_files_failed_total",
labels={"whisper_model": whisper_model, "api_name": api_name},
value=1
)
continue
temp_files.append(reencoded_mp3_path)
# Convert re-encoded MP3 to WAV
wav_file_path = convert_mp3_to_wav(reencoded_mp3_path)
if not os.path.exists(wav_file_path):
update_progress(f"Converted WAV file not found: {wav_file_path}")
failed_count += 1
log_counter(
metric_name="audio_files_failed_total",
labels={"whisper_model": whisper_model, "api_name": api_name},
value=1
)
continue
temp_files.append(wav_file_path)
# Initialize transcription
transcription = ""
# Transcribe audio
if diarize:
segments = speech_to_text(wav_file_path, whisper_model=whisper_model, diarize=True)
else:
segments = speech_to_text(wav_file_path, whisper_model=whisper_model)
# Handle segments nested under 'segments' key
if isinstance(segments, dict) and 'segments' in segments:
segments = segments['segments']
if isinstance(segments, list):
# Log first 5 segments for debugging
logging.debug(f"Segments before formatting: {segments[:5]}")
transcription = format_transcription_with_timestamps(segments)
logging.debug(f"Formatted transcription (first 500 chars): {transcription[:500]}")
update_progress("Audio transcribed successfully.")
else:
update_progress("Unexpected segments format received from speech_to_text.")
logging.error(f"Unexpected segments format: {segments}")
failed_count += 1
log_counter(
metric_name="audio_files_failed_total",
labels={"whisper_model": whisper_model, "api_name": api_name},
value=1
)
continue
if not transcription.strip():
update_progress("Transcription is empty.")
failed_count += 1
log_counter(
metric_name="audio_files_failed_total",
labels={"whisper_model": whisper_model, "api_name": api_name},
value=1
)
else:
# Apply chunking
chunked_text = improved_chunking_process(transcription, chunk_options)
# Summarize
logging.debug(f"Audio Transcription API Name: {api_name}")
if api_name:
try:
summary = perform_summarization(api_name, chunked_text, custom_prompt_input, api_key)
update_progress("Audio summarized successfully.")
except Exception as e:
logging.error(f"Error during summarization: {str(e)}")
summary = "Summary generation failed"
failed_count += 1
log_counter(
metric_name="audio_files_failed_total",
labels={"whisper_model": whisper_model, "api_name": api_name},
value=1
)
else:
summary = "No summary available (API not provided)"
all_transcriptions.append(transcription)
all_summaries.append(summary)
# Use custom_title if provided, otherwise use the original filename
title = custom_title if custom_title else os.path.basename(wav_file_path)
# Add to database
add_media_with_keywords(
url=url,
title=title,
media_type='audio',
content=transcription,
keywords=custom_keywords,
prompt=custom_prompt_input,
summary=summary,
transcription_model=whisper_model,
author="Unknown",
ingestion_date=datetime.now().strftime('%Y-%m-%d')
)
update_progress("Audio file processed and added to database.")
processed_count += 1
log_counter(
metric_name="audio_files_processed_total",
labels={"whisper_model": whisper_model, "api_name": api_name},
value=1
)
# Process uploaded file if provided
if audio_file:
url = generate_unique_id()
if os.path.getsize(audio_file.name) > MAX_FILE_SIZE:
update_progress(
f"Uploaded file size exceeds the maximum limit of {MAX_FILE_SIZE / (1024 * 1024):.2f}MB. Skipping this file.")
else:
try:
# Re-encode MP3 to fix potential issues
reencoded_mp3_path = reencode_mp3(audio_file.name)
if not os.path.exists(reencoded_mp3_path):
update_progress(f"Re-encoded file not found: {reencoded_mp3_path}")
return update_progress("Processing failed: Re-encoded file not found"), "", ""
temp_files.append(reencoded_mp3_path)
# Convert re-encoded MP3 to WAV
wav_file_path = convert_mp3_to_wav(reencoded_mp3_path)
if not os.path.exists(wav_file_path):
update_progress(f"Converted WAV file not found: {wav_file_path}")
return update_progress("Processing failed: Converted WAV file not found"), "", ""
temp_files.append(wav_file_path)
# Initialize transcription
transcription = ""
if diarize:
segments = speech_to_text(wav_file_path, whisper_model=whisper_model, diarize=True)
else:
segments = speech_to_text(wav_file_path, whisper_model=whisper_model)
# Handle segments nested under 'segments' key
if isinstance(segments, dict) and 'segments' in segments:
segments = segments['segments']
if isinstance(segments, list):
transcription = format_transcription_with_timestamps(segments)
else:
update_progress("Unexpected segments format received from speech_to_text.")
logging.error(f"Unexpected segments format: {segments}")
chunked_text = improved_chunking_process(transcription, chunk_options)
logging.debug(f"Audio Transcription API Name: {api_name}")
if api_name:
try:
summary = perform_summarization(api_name, chunked_text, custom_prompt_input, api_key)
update_progress("Audio summarized successfully.")
except Exception as e:
logging.error(f"Error during summarization: {str(e)}")
summary = "Summary generation failed"
else:
summary = "No summary available (API not provided)"
all_transcriptions.append(transcription)
all_summaries.append(summary)
# Use custom_title if provided, otherwise use the original filename
title = custom_title if custom_title else os.path.basename(wav_file_path)
add_media_with_keywords(
url="Uploaded File",
title=title,
media_type='audio',
content=transcription,
keywords=custom_keywords,
prompt=custom_prompt_input,
summary=summary,
transcription_model=whisper_model,
author="Unknown",
ingestion_date=datetime.now().strftime('%Y-%m-%d')
)
update_progress("Uploaded file processed and added to database.")
processed_count += 1
log_counter(
metric_name="audio_files_processed_total",
labels={"whisper_model": whisper_model, "api_name": api_name},
value=1
)
except Exception as e:
update_progress(f"Error processing uploaded file: {str(e)}")
logging.error(f"Error processing uploaded file: {str(e)}")
failed_count += 1
log_counter(
metric_name="audio_files_failed_total",
labels={"whisper_model": whisper_model, "api_name": api_name},
value=1
)
return update_progress("Processing failed: Error processing uploaded file"), "", ""
# Final cleanup
if not keep_original:
cleanup_files()
end_time = time.time()
processing_time = end_time - start_time
# Log processing time
log_histogram(
metric_name="audio_processing_time_seconds",
value=processing_time,
labels={"whisper_model": whisper_model, "api_name": api_name}
)
# Optionally, log total counts
log_counter(
metric_name="total_audio_files_processed",
labels={"whisper_model": whisper_model, "api_name": api_name},
value=processed_count
)
log_counter(
metric_name="total_audio_files_failed",
labels={"whisper_model": whisper_model, "api_name": api_name},
value=failed_count
)
final_progress = update_progress("All processing complete.")
final_transcriptions = "\n\n".join(all_transcriptions)
final_summaries = "\n\n".join(all_summaries)
return final_progress, final_transcriptions, final_summaries
except Exception as e:
logging.error(f"Error processing audio files: {str(e)}")
log_counter(
metric_name="audio_files_failed_total",
labels={"whisper_model": whisper_model, "api_name": api_name},
value=1
)
cleanup_files()
return update_progress(f"Processing failed: {str(e)}"), "", ""
def format_transcription_with_timestamps(segments, keep_timestamps):
"""
Formats the transcription segments with or without timestamps.
Parameters:
segments (list): List of transcription segments.
keep_timestamps (bool): Whether to include timestamps.
Returns:
str: Formatted transcription.
"""
if keep_timestamps:
formatted_segments = []
for segment in segments:
start = segment.get('Time_Start', 0)
end = segment.get('Time_End', 0)
text = segment.get('Text', '').strip()
formatted_segments.append(f"[{start:.2f}-{end:.2f}] {text}")
return "\n".join(formatted_segments)
else:
return "\n".join([segment.get('Text', '').strip() for segment in segments])
def download_youtube_audio(url):
try:
# Determine ffmpeg path based on the operating system.
ffmpeg_path = './Bin/ffmpeg.exe' if os.name == 'nt' else 'ffmpeg'
# Create a temporary directory
with tempfile.TemporaryDirectory() as temp_dir:
# Extract information about the video
with yt_dlp.YoutubeDL({'quiet': True}) as ydl:
info_dict = ydl.extract_info(url, download=False)
sanitized_title = sanitize_filename(info_dict['title'])
# Setup the temporary filenames
temp_video_path = Path(temp_dir) / f"{sanitized_title}_temp.mp4"
temp_audio_path = Path(temp_dir) / f"{sanitized_title}.mp3"
# Initialize yt-dlp with options for downloading
ydl_opts = {
'format': 'bestaudio[ext=m4a]/best[height<=480]', # Prefer best audio, or video up to 480p
'ffmpeg_location': ffmpeg_path,
'outtmpl': str(temp_video_path),
'noplaylist': True,
'quiet': True
}
# Execute yt-dlp to download the video/audio
with yt_dlp.YoutubeDL(ydl_opts) as ydl:
ydl.download([url])
# Check if the file exists
if not temp_video_path.exists():
raise FileNotFoundError(f"Expected file was not found: {temp_video_path}")
# Use ffmpeg to extract audio
ffmpeg_command = [
ffmpeg_path,
'-i', str(temp_video_path),
'-vn', # No video
'-acodec', 'libmp3lame',
'-b:a', '192k',
str(temp_audio_path)
]
subprocess.run(ffmpeg_command, check=True, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
# Check if the audio file was created
if not temp_audio_path.exists():
raise FileNotFoundError(f"Expected audio file was not found: {temp_audio_path}")
# Create a persistent directory for the download if it doesn't exist
persistent_dir = Path("downloads")
persistent_dir.mkdir(exist_ok=True)
# Move the file from the temporary directory to the persistent directory
persistent_file_path = persistent_dir / f"{sanitized_title}.mp3"
os.replace(str(temp_audio_path), str(persistent_file_path))
# Add the file to the list of downloaded files
downloaded_files.append(str(persistent_file_path))
return str(persistent_file_path), f"Audio downloaded successfully: {sanitized_title}.mp3"
except Exception as e:
return None, f"Error downloading audio: {str(e)}"
def process_podcast(url, title, author, keywords, custom_prompt, api_name, api_key, whisper_model,
keep_original=False, enable_diarization=False, use_cookies=False, cookies=None,
chunk_method=None, max_chunk_size=300, chunk_overlap=0, use_adaptive_chunking=False,
use_multi_level_chunking=False, chunk_language='english', keep_timestamps=True):
"""
Processes a podcast by downloading the audio, transcribing it, summarizing the transcription,
and adding the results to the database. Metrics are logged throughout the process.
Parameters:
url (str): URL of the podcast.
title (str): Title of the podcast.
author (str): Author of the podcast.
keywords (str): Comma-separated keywords.
custom_prompt (str): Custom prompt for summarization.
api_name (str): API name for summarization.
api_key (str): API key for summarization.
whisper_model (str): Whisper model to use for transcription.
keep_original (bool): Whether to keep the original audio file.
enable_diarization (bool): Whether to enable speaker diarization.
use_cookies (bool): Whether to use cookies for authenticated downloads.
cookies (str): JSON-formatted cookies string.
chunk_method (str): Method for chunking text.
max_chunk_size (int): Maximum size for each text chunk.
chunk_overlap (int): Overlap size between chunks.
use_adaptive_chunking (bool): Whether to use adaptive chunking.
use_multi_level_chunking (bool): Whether to use multi-level chunking.
chunk_language (str): Language for chunking.
keep_timestamps (bool): Whether to keep timestamps in transcription.
Returns:
tuple: (progress_message, transcription, summary, title, author, keywords, error_message)
"""
start_time = time.time() # Start time for processing
error_message = ""
temp_files = []
# Define labels for metrics
labels = {
"whisper_model": whisper_model,
"api_name": api_name if api_name else "None"
}
def update_progress(message):
"""
Updates the progress messages.
Parameters:
message (str): Progress message to append.
Returns:
str: Combined progress messages.
"""
progress.append(message)
return "\n".join(progress)
def cleanup_files():
if not keep_original:
for file in temp_files:
try:
if os.path.exists(file):
os.remove(file)
update_progress(f"Temporary file {file} removed.")
except Exception as e:
update_progress(f"Failed to remove temporary file {file}: {str(e)}")
progress = [] # Initialize progress messages
try:
# Handle cookies if required
if use_cookies:
cookies = json.loads(cookies)
# Download the podcast audio file
audio_file = download_audio_file(url, whisper_model, use_cookies, cookies)
if not audio_file:
raise RuntimeError("Failed to download podcast audio.")
temp_files.append(audio_file)
update_progress("Podcast downloaded successfully.")
# Extract metadata from the podcast
metadata = extract_metadata(url)
title = title or metadata.get('title', 'Unknown Podcast')
author = author or metadata.get('uploader', 'Unknown Author')
# Format metadata for storage
metadata_text = f"""
Metadata:
Title: {title}
Author: {author}
Series: {metadata.get('series', 'N/A')}
Episode: {metadata.get('episode', 'N/A')}
Season: {metadata.get('season', 'N/A')}
Upload Date: {metadata.get('upload_date', 'N/A')}
Duration: {metadata.get('duration', 'N/A')} seconds
Description: {metadata.get('description', 'N/A')}
"""
# Update keywords with metadata information
new_keywords = []
if metadata.get('series'):
new_keywords.append(f"series:{metadata['series']}")
if metadata.get('episode'):
new_keywords.append(f"episode:{metadata['episode']}")
if metadata.get('season'):
new_keywords.append(f"season:{metadata['season']}")
keywords = f"{keywords},{','.join(new_keywords)}" if keywords else ','.join(new_keywords)
update_progress(f"Metadata extracted - Title: {title}, Author: {author}, Keywords: {keywords}")
# Transcribe the podcast audio
try:
if enable_diarization:
segments = speech_to_text(audio_file, whisper_model=whisper_model, diarize=True)
else:
segments = speech_to_text(audio_file, whisper_model=whisper_model)
# SEems like this could be optimized... FIXME
def format_segment(segment):
start = segment.get('start', 0)
end = segment.get('end', 0)
text = segment.get('Text', '')
if isinstance(segments, dict) and 'segments' in segments:
segments = segments['segments']
if isinstance(segments, list):
transcription = format_transcription_with_timestamps(segments, keep_timestamps)
update_progress("Podcast transcribed successfully.")
else:
raise ValueError("Unexpected segments format received from speech_to_text.")
if not transcription.strip():
raise ValueError("Transcription is empty.")
except Exception as e:
error_message = f"Transcription failed: {str(e)}"
raise RuntimeError(error_message)
# Apply chunking to the transcription
chunk_options = {
'method': chunk_method,
'max_size': max_chunk_size,
'overlap': chunk_overlap,
'adaptive': use_adaptive_chunking,
'multi_level': use_multi_level_chunking,
'language': chunk_language
}
chunked_text = improved_chunking_process(transcription, chunk_options)
# Combine metadata and transcription
full_content = metadata_text + "\n\nTranscription:\n" + transcription
# Summarize the transcription if API is provided
summary = None
if api_name:
try:
summary = perform_summarization(api_name, chunked_text, custom_prompt, api_key)
update_progress("Podcast summarized successfully.")
except Exception as e:
error_message = f"Summarization failed: {str(e)}"
raise RuntimeError(error_message)
else:
summary = "No summary available (API not provided)"
# Add the processed podcast to the database
try:
add_media_with_keywords(
url=url,
title=title,
media_type='podcast',
content=full_content,
keywords=keywords,
prompt=custom_prompt,
summary=summary or "No summary available",
transcription_model=whisper_model,
author=author,
ingestion_date=datetime.now().strftime('%Y-%m-%d')
)
update_progress("Podcast added to database successfully.")
except Exception as e:
error_message = f"Error adding podcast to database: {str(e)}"
raise RuntimeError(error_message)
# Cleanup temporary files if required
cleanup_files()
# Calculate processing time
end_time = time.time()
processing_time = end_time - start_time
# Log successful processing
log_counter(
metric_name="podcasts_processed_total",
labels=labels,
value=1
)
# Log processing time
log_histogram(
metric_name="podcast_processing_time_seconds",
value=processing_time,
labels=labels
)
# Return the final outputs
final_progress = update_progress("Processing complete.")
return (final_progress, full_content, summary or "No summary generated.",
title, author, keywords, error_message)
except Exception as e:
# Calculate processing time up to the point of failure
end_time = time.time()
processing_time = end_time - start_time
# Log failed processing
log_counter(
metric_name="podcasts_failed_total",
labels=labels,
value=1
)
# Log processing time even on failure
log_histogram(
metric_name="podcast_processing_time_seconds",
value=processing_time,
labels=labels
)
logging.error(f"Error processing podcast: {str(e)}")
cleanup_files()
final_progress = update_progress(f"Processing failed: {str(e)}")
return (final_progress, "", "", "", "", "", str(e))
#
#
#######################################################################################################################