codebook / potato /server_utils /waveform_service.py
davidjurgens's picture
Deploy: Potato — Codebook Annotation
aceb1b2 verified
Raw
History Blame Contribute Delete
18.3 kB
"""
Waveform Service
Handles generation and caching of audio waveform data for the audio annotation feature.
Uses BBC's audiowaveform tool to generate pre-computed waveform data files.
Features:
- LRU cache for waveform files
- Background look-ahead pre-computation for upcoming instances
- Support for both local files and URLs
- Graceful fallback if audiowaveform not installed
"""
import os
import logging
import hashlib
import subprocess
import shutil
import tempfile
import threading
import time
from typing import Optional, List, Dict
from collections import OrderedDict
from urllib.parse import urlparse
from pathlib import Path
try:
import requests
REQUESTS_AVAILABLE = True
except ImportError:
REQUESTS_AVAILABLE = False
logger = logging.getLogger(__name__)
class WaveformService:
"""
Service for generating and caching audio waveform data.
Uses BBC's audiowaveform tool to generate pre-computed waveform data
that can be efficiently rendered by Peaks.js on the frontend.
"""
# Default configuration
DEFAULT_LOOK_AHEAD = 5
DEFAULT_CACHE_MAX_SIZE = 100
DEFAULT_CLIENT_FALLBACK_MAX_DURATION = 1800 # 30 minutes in seconds
# Waveform generation settings
WAVEFORM_ZOOM_LEVEL = 256 # Samples per pixel
WAVEFORM_BITS = 8 # 8-bit resolution
def __init__(
self,
cache_dir: str,
look_ahead: int = DEFAULT_LOOK_AHEAD,
cache_max_size: int = DEFAULT_CACHE_MAX_SIZE,
client_fallback_max_duration: int = DEFAULT_CLIENT_FALLBACK_MAX_DURATION
):
"""
Initialize the WaveformService.
Args:
cache_dir: Directory to store generated waveform files
look_ahead: Number of instances to pre-compute ahead
cache_max_size: Maximum number of cached waveform files
client_fallback_max_duration: Max duration (seconds) for client-side fallback
"""
self.cache_dir = cache_dir
self.look_ahead = look_ahead
self.cache_max_size = cache_max_size
self.client_fallback_max_duration = client_fallback_max_duration
# LRU cache tracking
self._cache_order: OrderedDict = OrderedDict()
self._cache_lock = threading.Lock()
# Background pre-computation
self._precompute_thread: Optional[threading.Thread] = None
self._precompute_queue: List[str] = []
self._precompute_lock = threading.Lock()
self._stop_precompute = threading.Event()
# Check if audiowaveform is installed
self._audiowaveform_available = self._check_audiowaveform_installed()
# Ensure cache directory exists
self._ensure_cache_dir()
logger.info(f"WaveformService initialized: cache_dir={cache_dir}, "
f"look_ahead={look_ahead}, audiowaveform_available={self._audiowaveform_available}")
def _ensure_cache_dir(self) -> None:
"""Create the cache directory if it doesn't exist."""
if not os.path.exists(self.cache_dir):
os.makedirs(self.cache_dir, exist_ok=True)
logger.info(f"Created waveform cache directory: {self.cache_dir}")
def _check_audiowaveform_installed(self) -> bool:
"""
Check if the audiowaveform tool is installed and available.
Returns:
True if audiowaveform is available, False otherwise
"""
try:
result = subprocess.run(
['audiowaveform', '--version'],
capture_output=True,
text=True,
timeout=5
)
if result.returncode == 0:
version = result.stdout.strip() or result.stderr.strip()
logger.info(f"audiowaveform found: {version}")
return True
except (subprocess.SubprocessError, FileNotFoundError, OSError) as e:
logger.warning(f"audiowaveform not available: {e}")
return False
@property
def is_available(self) -> bool:
"""Check if waveform generation is available."""
return self._audiowaveform_available
def _get_cache_key(self, audio_path: str) -> str:
"""
Generate a unique cache key for an audio file.
Args:
audio_path: Path or URL to the audio file
Returns:
MD5 hash of the path as cache key
"""
return hashlib.md5(audio_path.encode('utf-8')).hexdigest()
def _get_waveform_cache_path(self, audio_path: str) -> str:
"""
Get the cache file path for a waveform.
Args:
audio_path: Path or URL to the audio file
Returns:
Full path to the waveform cache file
"""
cache_key = self._get_cache_key(audio_path)
return os.path.join(self.cache_dir, f"{cache_key}.dat")
def _is_url(self, path: str) -> bool:
"""
Check if a path is a URL.
Args:
path: The path to check
Returns:
True if path is a URL, False otherwise
"""
return path.startswith(('http://', 'https://', '//'))
def _download_audio(self, url: str) -> Optional[str]:
"""
Download an audio file from URL to a temporary file.
Args:
url: URL of the audio file
Returns:
Path to temporary file, or None if download failed
"""
if not REQUESTS_AVAILABLE:
logger.error("requests library not available for downloading audio")
return None
try:
# Determine file extension from URL
parsed = urlparse(url)
path = parsed.path
ext = os.path.splitext(path)[1] or '.mp3'
# Create temporary file
temp_fd, temp_path = tempfile.mkstemp(suffix=ext)
os.close(temp_fd)
logger.debug(f"Downloading audio from {url} to {temp_path}")
response = requests.get(url, stream=True, timeout=60)
response.raise_for_status()
with open(temp_path, 'wb') as f:
for chunk in response.iter_content(chunk_size=8192):
f.write(chunk)
logger.debug(f"Downloaded audio: {os.path.getsize(temp_path)} bytes")
return temp_path
except Exception as e:
logger.error(f"Failed to download audio from {url}: {e}")
return None
def _generate_waveform(self, audio_path: str, output_path: str) -> bool:
"""
Generate waveform data using audiowaveform tool.
Args:
audio_path: Path to the audio file (local)
output_path: Path to write the waveform data file
Returns:
True if generation succeeded, False otherwise
"""
if not self._audiowaveform_available:
logger.warning("audiowaveform not available, cannot generate waveform")
return False
try:
# Build command
cmd = [
'audiowaveform',
'-i', audio_path,
'-o', output_path,
'-z', str(self.WAVEFORM_ZOOM_LEVEL),
'-b', str(self.WAVEFORM_BITS),
]
logger.debug(f"Running: {' '.join(cmd)}")
result = subprocess.run(
cmd,
capture_output=True,
text=True,
timeout=300 # 5 minute timeout for long files
)
if result.returncode == 0:
logger.info(f"Generated waveform: {output_path}")
return True
else:
logger.error(f"audiowaveform failed: {result.stderr}")
return False
except subprocess.TimeoutExpired:
logger.error(f"audiowaveform timed out for {audio_path}")
return False
except Exception as e:
logger.error(f"Error generating waveform for {audio_path}: {e}")
return False
def _update_cache_order(self, cache_path: str) -> None:
"""
Update LRU cache order and evict if necessary.
Args:
cache_path: Path to the cache file being accessed
"""
with self._cache_lock:
# Move to end (most recently used)
if cache_path in self._cache_order:
self._cache_order.move_to_end(cache_path)
else:
self._cache_order[cache_path] = True
# Evict oldest if over limit
while len(self._cache_order) > self.cache_max_size:
oldest_path, _ = self._cache_order.popitem(last=False)
if os.path.exists(oldest_path):
try:
os.remove(oldest_path)
logger.debug(f"Evicted from cache: {oldest_path}")
except OSError as e:
logger.warning(f"Failed to remove cache file {oldest_path}: {e}")
def get_waveform_path(self, audio_path: str, generate: bool = True) -> Optional[str]:
"""
Get the waveform data file path for an audio file.
If the waveform doesn't exist and generate=True, it will be generated.
Args:
audio_path: Path or URL to the audio file
generate: Whether to generate if not cached
Returns:
Path to waveform data file, or None if not available
"""
cache_path = self._get_waveform_cache_path(audio_path)
# Check if already cached
if os.path.exists(cache_path):
self._update_cache_order(cache_path)
logger.debug(f"Waveform cache hit: {cache_path}")
return cache_path
if not generate:
return None
# Generate waveform
temp_audio = None
try:
# Handle URL vs local path
if self._is_url(audio_path):
temp_audio = self._download_audio(audio_path)
if not temp_audio:
return None
local_path = temp_audio
else:
local_path = audio_path
if not os.path.exists(local_path):
logger.warning(f"Audio file not found: {local_path}")
return None
# Generate waveform
if self._generate_waveform(local_path, cache_path):
self._update_cache_order(cache_path)
return cache_path
else:
return None
finally:
# Clean up temporary file
if temp_audio and os.path.exists(temp_audio):
try:
os.remove(temp_audio)
except OSError:
pass
def get_waveform_url(self, audio_path: str, base_url: str = '/api/waveform/') -> Optional[str]:
"""
Get the URL to fetch waveform data for an audio file.
Args:
audio_path: Path or URL to the audio file
base_url: Base URL for the waveform API endpoint
Returns:
URL to fetch waveform data
"""
cache_key = self._get_cache_key(audio_path)
return f"{base_url}{cache_key}"
def precompute_batch(self, audio_paths: List[str]) -> None:
"""
Pre-compute waveforms for a batch of audio files.
This is called synchronously and blocks until all are complete.
Use start_background_precompute for non-blocking operation.
Args:
audio_paths: List of audio file paths or URLs
"""
for audio_path in audio_paths:
if audio_path:
self.get_waveform_path(audio_path, generate=True)
def queue_precompute(self, audio_paths: List[str]) -> None:
"""
Add audio files to the background pre-computation queue.
Args:
audio_paths: List of audio file paths or URLs to pre-compute
"""
with self._precompute_lock:
# Only add paths not already in queue or cached
for path in audio_paths:
if path and path not in self._precompute_queue:
cache_path = self._get_waveform_cache_path(path)
if not os.path.exists(cache_path):
self._precompute_queue.append(path)
# Start background thread if not running
if self._precompute_thread is None or not self._precompute_thread.is_alive():
self._start_background_precompute()
def _start_background_precompute(self) -> None:
"""Start the background pre-computation thread."""
self._stop_precompute.clear()
self._precompute_thread = threading.Thread(
target=self._background_precompute_worker,
daemon=True
)
self._precompute_thread.start()
logger.debug("Started background waveform pre-computation thread")
def _background_precompute_worker(self) -> None:
"""Background worker for pre-computing waveforms."""
while not self._stop_precompute.is_set():
# Get next item from queue
audio_path = None
with self._precompute_lock:
if self._precompute_queue:
audio_path = self._precompute_queue.pop(0)
if audio_path:
logger.debug(f"Background pre-computing waveform for: {audio_path}")
self.get_waveform_path(audio_path, generate=True)
else:
# No more items, exit thread
break
# Small delay between items to avoid overloading
time.sleep(0.1)
logger.debug("Background waveform pre-computation thread finished")
def stop_background_precompute(self) -> None:
"""Stop the background pre-computation thread."""
self._stop_precompute.set()
if self._precompute_thread and self._precompute_thread.is_alive():
self._precompute_thread.join(timeout=5)
def get_audio_duration(self, audio_path: str) -> Optional[float]:
"""
Get the duration of an audio file in seconds.
Uses ffprobe if available, otherwise returns None.
Args:
audio_path: Path to the audio file
Returns:
Duration in seconds, or None if cannot determine
"""
try:
result = subprocess.run(
[
'ffprobe',
'-v', 'error',
'-show_entries', 'format=duration',
'-of', 'default=noprint_wrappers=1:nokey=1',
audio_path
],
capture_output=True,
text=True,
timeout=10
)
if result.returncode == 0:
return float(result.stdout.strip())
except (subprocess.SubprocessError, ValueError, FileNotFoundError):
pass
return None
def should_use_client_fallback(self, audio_path: str) -> bool:
"""
Determine if client-side waveform generation should be used.
Client-side is preferred for short files when server-side is not available.
Args:
audio_path: Path to the audio file
Returns:
True if client-side fallback should be used
"""
if self._audiowaveform_available:
return False
duration = self.get_audio_duration(audio_path)
if duration is not None and duration <= self.client_fallback_max_duration:
return True
return False
def clear_cache(self) -> int:
"""
Clear all cached waveform files.
Returns:
Number of files removed
"""
count = 0
with self._cache_lock:
for cache_path in list(self._cache_order.keys()):
if os.path.exists(cache_path):
try:
os.remove(cache_path)
count += 1
except OSError as e:
logger.warning(f"Failed to remove {cache_path}: {e}")
self._cache_order.clear()
logger.info(f"Cleared {count} cached waveform files")
return count
def get_cache_stats(self) -> Dict:
"""
Get statistics about the waveform cache.
Returns:
Dictionary with cache statistics
"""
with self._cache_lock:
cached_files = len(self._cache_order)
total_size = 0
for cache_path in self._cache_order.keys():
if os.path.exists(cache_path):
total_size += os.path.getsize(cache_path)
return {
'cached_files': cached_files,
'max_files': self.cache_max_size,
'total_size_bytes': total_size,
'total_size_mb': round(total_size / (1024 * 1024), 2),
'cache_dir': self.cache_dir,
'audiowaveform_available': self._audiowaveform_available,
}
# Global instance (initialized when needed)
_waveform_service: Optional[WaveformService] = None
def get_waveform_service() -> Optional[WaveformService]:
"""Get the global WaveformService instance."""
return _waveform_service
def init_waveform_service(
cache_dir: str,
look_ahead: int = WaveformService.DEFAULT_LOOK_AHEAD,
cache_max_size: int = WaveformService.DEFAULT_CACHE_MAX_SIZE,
client_fallback_max_duration: int = WaveformService.DEFAULT_CLIENT_FALLBACK_MAX_DURATION
) -> WaveformService:
"""
Initialize the global WaveformService instance.
Args:
cache_dir: Directory to store generated waveform files
look_ahead: Number of instances to pre-compute ahead
cache_max_size: Maximum number of cached waveform files
client_fallback_max_duration: Max duration for client-side fallback
Returns:
The initialized WaveformService instance
"""
global _waveform_service
_waveform_service = WaveformService(
cache_dir=cache_dir,
look_ahead=look_ahead,
cache_max_size=cache_max_size,
client_fallback_max_duration=client_fallback_max_duration
)
return _waveform_service