580 lines
22 KiB
Python
580 lines
22 KiB
Python
"""Sound normalizer service for normalizing audio files using ffmpeg loudnorm."""
|
|
|
|
import hashlib
|
|
import json
|
|
import os
|
|
import re
|
|
from pathlib import Path
|
|
from typing import TypedDict
|
|
|
|
import ffmpeg # type: ignore[import-untyped]
|
|
from sqlmodel.ext.asyncio.session import AsyncSession
|
|
|
|
from app.core.config import settings
|
|
from app.core.logging import get_logger
|
|
from app.models.sound import Sound
|
|
from app.repositories.sound import SoundRepository
|
|
|
|
logger = get_logger(__name__)
|
|
|
|
|
|
class NormalizationInfo(TypedDict):
|
|
"""Type definition for normalization information in results."""
|
|
|
|
filename: str
|
|
status: str
|
|
reason: str | None
|
|
original_path: str | None
|
|
normalized_path: str | None
|
|
normalized_filename: str | None
|
|
normalized_duration: int | None
|
|
normalized_size: int | None
|
|
normalized_hash: str | None
|
|
id: int | None
|
|
error: str | None
|
|
|
|
|
|
class NormalizationResults(TypedDict):
|
|
"""Type definition for normalization results."""
|
|
|
|
processed: int
|
|
normalized: int
|
|
skipped: int
|
|
errors: int
|
|
files: list[NormalizationInfo]
|
|
|
|
|
|
class SoundNormalizerService:
|
|
"""Service for normalizing audio files using ffmpeg loudnorm."""
|
|
|
|
def __init__(self, session: AsyncSession) -> None:
|
|
"""Initialize the sound normalizer service."""
|
|
self.session = session
|
|
self.sound_repo = SoundRepository(session)
|
|
|
|
# Normalization settings from config
|
|
self.output_format = settings.NORMALIZED_AUDIO_FORMAT
|
|
self.output_bitrate = settings.NORMALIZED_AUDIO_BITRATE
|
|
self.passes = settings.NORMALIZED_AUDIO_PASSES
|
|
|
|
# Directory mappings for different sound types
|
|
self.type_directories = {
|
|
"SDB": "sounds/normalized/soundboard",
|
|
"TTS": "sounds/normalized/text_to_speech",
|
|
"EXT": "sounds/normalized/extracted",
|
|
}
|
|
|
|
# Ensure normalized directories exist
|
|
self._ensure_directories()
|
|
|
|
def _ensure_directories(self) -> None:
|
|
"""Ensure all normalized sound directories exist."""
|
|
for directory in self.type_directories.values():
|
|
Path(directory).mkdir(parents=True, exist_ok=True)
|
|
logger.debug("Ensured directory exists: %s", directory)
|
|
|
|
def _get_normalized_path(self, sound: Sound) -> Path:
|
|
"""Get the normalized file path for a sound."""
|
|
return self._get_normalized_path_from_data(sound.type, sound.filename)
|
|
|
|
def _get_normalized_path_from_data(self, sound_type: str, filename: str) -> Path:
|
|
"""Get the normalized file path from sound data."""
|
|
# Get the appropriate directory for the sound type
|
|
directory = self.type_directories.get(sound_type, "sounds/normalized/other")
|
|
|
|
# Create the directory if it doesn't exist
|
|
Path(directory).mkdir(parents=True, exist_ok=True)
|
|
|
|
# Generate filename: original_name.{format}
|
|
original_stem = Path(filename).stem
|
|
normalized_filename = f"{original_stem}.{self.output_format}"
|
|
|
|
return Path(directory) / normalized_filename
|
|
|
|
def _get_original_path(self, sound: Sound) -> Path:
|
|
"""Get the original file path for a sound."""
|
|
return self._get_original_path_from_data(sound.type, sound.filename)
|
|
|
|
def _get_original_path_from_data(self, sound_type: str, filename: str) -> Path:
|
|
"""Get the original file path from sound data."""
|
|
# Map sound types to their original directories
|
|
type_to_original_dir = {
|
|
"SDB": "sounds/originals/soundboard",
|
|
"TTS": "sounds/originals/text_to_speech",
|
|
"EXT": "sounds/originals/extracted",
|
|
}
|
|
|
|
original_dir = type_to_original_dir.get(sound_type, "sounds/originals/other")
|
|
return Path(original_dir) / filename
|
|
|
|
def _get_file_hash(self, file_path: Path) -> str:
|
|
"""Calculate SHA-256 hash of a file."""
|
|
hash_sha256 = hashlib.sha256()
|
|
with open(file_path, "rb") as f:
|
|
for chunk in iter(lambda: f.read(4096), b""):
|
|
hash_sha256.update(chunk)
|
|
return hash_sha256.hexdigest()
|
|
|
|
def _get_file_size(self, file_path: Path) -> int:
|
|
"""Get file size in bytes."""
|
|
return file_path.stat().st_size
|
|
|
|
def _get_audio_duration(self, file_path: Path) -> int:
|
|
"""Get audio duration in milliseconds using ffmpeg."""
|
|
try:
|
|
probe = ffmpeg.probe(str(file_path))
|
|
duration = float(probe["format"]["duration"])
|
|
return int(duration * 1000) # Convert to milliseconds
|
|
except Exception as e:
|
|
logger.warning("Failed to get duration for %s: %s", file_path, e)
|
|
return 0
|
|
|
|
async def _normalize_audio_one_pass(
|
|
self,
|
|
input_path: Path,
|
|
output_path: Path,
|
|
) -> None:
|
|
"""Normalize audio using one-pass loudnorm."""
|
|
try:
|
|
logger.info(
|
|
"Starting one-pass normalization: %s -> %s",
|
|
input_path,
|
|
output_path,
|
|
)
|
|
|
|
stream = ffmpeg.input(str(input_path))
|
|
stream = ffmpeg.filter(stream, "loudnorm", I=-23, TP=-2, LRA=7)
|
|
|
|
# Apply output format and bitrate
|
|
output_args = {}
|
|
if self.output_format == "mp3":
|
|
output_args["acodec"] = "libmp3lame"
|
|
output_args["audio_bitrate"] = self.output_bitrate
|
|
elif self.output_format == "aac":
|
|
output_args["acodec"] = "aac"
|
|
output_args["audio_bitrate"] = self.output_bitrate
|
|
elif self.output_format == "opus":
|
|
output_args["acodec"] = "libopus"
|
|
output_args["audio_bitrate"] = self.output_bitrate
|
|
|
|
stream = ffmpeg.output(stream, str(output_path), **output_args)
|
|
stream = ffmpeg.overwrite_output(stream)
|
|
|
|
ffmpeg.run(stream, quiet=True, overwrite_output=True)
|
|
logger.info("One-pass normalization completed: %s", output_path)
|
|
|
|
except Exception as e:
|
|
logger.exception("One-pass normalization failed for %s", input_path)
|
|
raise
|
|
|
|
async def _normalize_audio_two_pass(
|
|
self,
|
|
input_path: Path,
|
|
output_path: Path,
|
|
) -> None:
|
|
"""Normalize audio using two-pass loudnorm for better quality."""
|
|
try:
|
|
logger.info(
|
|
"Starting two-pass normalization: %s -> %s", input_path, output_path
|
|
)
|
|
|
|
# First pass: analyze
|
|
logger.debug("First pass: analyzing %s", input_path)
|
|
|
|
stream = ffmpeg.input(str(input_path))
|
|
stream = ffmpeg.filter(
|
|
stream,
|
|
"loudnorm",
|
|
I=-23,
|
|
TP=-2,
|
|
LRA=7,
|
|
print_format="json",
|
|
)
|
|
# Output to null device with explicit format
|
|
null_output = "/dev/null" if os.name != "nt" else "NUL"
|
|
stream = ffmpeg.output(stream, null_output, format="null")
|
|
|
|
# Run first pass and capture output
|
|
try:
|
|
result = ffmpeg.run(stream, capture_stderr=True, quiet=True)
|
|
analysis_output = result[1].decode("utf-8")
|
|
except ffmpeg.Error as e:
|
|
logger.error("FFmpeg first pass failed for %s. Stdout: %s, Stderr: %s",
|
|
input_path, e.stdout.decode() if e.stdout else "None",
|
|
e.stderr.decode() if e.stderr else "None")
|
|
raise
|
|
|
|
# Extract loudnorm measurements from the output
|
|
# The JSON output is at the end of stderr
|
|
logger.debug("Loudnorm analysis output: %s", analysis_output)
|
|
|
|
# Find JSON in the output
|
|
json_match = re.search(r'\{[^{}]*"input_i"[^{}]*\}', analysis_output)
|
|
if not json_match:
|
|
logger.error("Could not find JSON in loudnorm output: %s", analysis_output)
|
|
raise ValueError("Could not extract loudnorm analysis data")
|
|
|
|
logger.debug("Found JSON match: %s", json_match.group())
|
|
analysis_data = json.loads(json_match.group())
|
|
|
|
# Check for invalid values that would cause second pass to fail
|
|
invalid_values = ["-inf", "inf", "nan"]
|
|
for key in ["input_i", "input_lra", "input_tp", "input_thresh", "target_offset"]:
|
|
if str(analysis_data.get(key, "")).lower() in invalid_values:
|
|
logger.warning(
|
|
"Invalid analysis value for %s: %s. Falling back to one-pass normalization.",
|
|
key, analysis_data.get(key)
|
|
)
|
|
# Fall back to one-pass normalization
|
|
await self._normalize_audio_one_pass(input_path, output_path)
|
|
return
|
|
|
|
# Second pass: normalize with measured values
|
|
logger.debug("Second pass: normalizing %s with measured values", input_path)
|
|
|
|
stream = ffmpeg.input(str(input_path))
|
|
stream = ffmpeg.filter(
|
|
stream,
|
|
"loudnorm",
|
|
measured_I=analysis_data["input_i"],
|
|
measured_LRA=analysis_data["input_lra"],
|
|
measured_TP=analysis_data["input_tp"],
|
|
measured_thresh=analysis_data["input_thresh"],
|
|
offset=analysis_data["target_offset"],
|
|
)
|
|
|
|
# Apply output format and bitrate
|
|
output_args = {}
|
|
if self.output_format == "mp3":
|
|
output_args["acodec"] = "libmp3lame"
|
|
output_args["audio_bitrate"] = self.output_bitrate
|
|
elif self.output_format == "aac":
|
|
output_args["acodec"] = "aac"
|
|
output_args["audio_bitrate"] = self.output_bitrate
|
|
elif self.output_format == "opus":
|
|
output_args["acodec"] = "libopus"
|
|
output_args["audio_bitrate"] = self.output_bitrate
|
|
|
|
stream = ffmpeg.output(stream, str(output_path), **output_args)
|
|
stream = ffmpeg.overwrite_output(stream)
|
|
|
|
try:
|
|
ffmpeg.run(stream, quiet=True, overwrite_output=True)
|
|
logger.info("Two-pass normalization completed: %s", output_path)
|
|
except ffmpeg.Error as e:
|
|
logger.error("FFmpeg second pass failed for %s. Stdout: %s, Stderr: %s",
|
|
input_path, e.stdout.decode() if e.stdout else "None",
|
|
e.stderr.decode() if e.stderr else "None")
|
|
raise
|
|
|
|
except Exception as e:
|
|
logger.exception("Two-pass normalization failed for %s", input_path)
|
|
raise
|
|
|
|
async def normalize_sound(
|
|
self,
|
|
sound: Sound,
|
|
force: bool = False,
|
|
one_pass: bool | None = None,
|
|
sound_data: dict | None = None,
|
|
) -> NormalizationInfo:
|
|
"""Normalize a single sound."""
|
|
# Use provided sound_data to avoid detached instance issues, or capture from sound
|
|
if sound_data:
|
|
filename = sound_data["filename"]
|
|
sound_id = sound_data["id"]
|
|
is_normalized = sound_data["is_normalized"]
|
|
sound_type = sound_data["type"]
|
|
else:
|
|
# Fallback to accessing sound properties (for single sound normalization)
|
|
filename = sound.filename
|
|
sound_id = sound.id
|
|
is_normalized = sound.is_normalized
|
|
sound_type = sound.type
|
|
|
|
# Check if already normalized and not forcing
|
|
if is_normalized and not force:
|
|
return {
|
|
"filename": filename,
|
|
"status": "skipped",
|
|
"reason": "already normalized",
|
|
"original_path": None,
|
|
"normalized_path": None,
|
|
"normalized_filename": None,
|
|
"normalized_duration": None,
|
|
"normalized_size": None,
|
|
"normalized_hash": None,
|
|
"id": sound_id,
|
|
"error": None,
|
|
}
|
|
|
|
try:
|
|
# Get paths using captured data to avoid accessing sound properties
|
|
original_path = self._get_original_path_from_data(sound_type, filename)
|
|
normalized_path = self._get_normalized_path_from_data(sound_type, filename)
|
|
|
|
# Check if original file exists
|
|
if not original_path.exists():
|
|
error_msg = f"Original file not found: {original_path}"
|
|
logger.error(error_msg)
|
|
return {
|
|
"filename": filename,
|
|
"status": "error",
|
|
"reason": None,
|
|
"original_path": str(original_path),
|
|
"normalized_path": None,
|
|
"normalized_filename": None,
|
|
"normalized_duration": None,
|
|
"normalized_size": None,
|
|
"normalized_hash": None,
|
|
"id": sound_id,
|
|
"error": error_msg,
|
|
}
|
|
|
|
# Determine which normalization method to use
|
|
use_one_pass = one_pass if one_pass is not None else (self.passes == 1)
|
|
|
|
# Perform normalization
|
|
if use_one_pass:
|
|
await self._normalize_audio_one_pass(original_path, normalized_path)
|
|
else:
|
|
await self._normalize_audio_two_pass(original_path, normalized_path)
|
|
|
|
# Get normalized file info
|
|
normalized_duration = self._get_audio_duration(normalized_path)
|
|
normalized_size = self._get_file_size(normalized_path)
|
|
normalized_hash = self._get_file_hash(normalized_path)
|
|
normalized_filename = normalized_path.name
|
|
|
|
# Update sound in database
|
|
update_data = {
|
|
"normalized_filename": normalized_filename,
|
|
"normalized_duration": normalized_duration,
|
|
"normalized_size": normalized_size,
|
|
"normalized_hash": normalized_hash,
|
|
"is_normalized": True,
|
|
}
|
|
|
|
await self.sound_repo.update(sound, update_data)
|
|
logger.info("Normalized sound: %s -> %s", filename, normalized_filename)
|
|
|
|
return {
|
|
"filename": filename,
|
|
"status": "normalized",
|
|
"reason": None,
|
|
"original_path": str(original_path),
|
|
"normalized_path": str(normalized_path),
|
|
"normalized_filename": normalized_filename,
|
|
"normalized_duration": normalized_duration,
|
|
"normalized_size": normalized_size,
|
|
"normalized_hash": normalized_hash,
|
|
"id": sound_id,
|
|
"error": None,
|
|
}
|
|
|
|
except Exception as e:
|
|
error_msg = str(e)
|
|
logger.exception(
|
|
"Failed to normalize sound %s",
|
|
filename,
|
|
)
|
|
return {
|
|
"filename": filename,
|
|
"status": "error",
|
|
"reason": None,
|
|
"original_path": (
|
|
str(original_path) if "original_path" in locals() else None
|
|
),
|
|
"normalized_path": (
|
|
str(normalized_path) if "normalized_path" in locals() else None
|
|
),
|
|
"normalized_filename": None,
|
|
"normalized_duration": None,
|
|
"normalized_size": None,
|
|
"normalized_hash": None,
|
|
"id": sound_id,
|
|
"error": error_msg,
|
|
}
|
|
|
|
async def normalize_all_sounds(
|
|
self,
|
|
force: bool = False,
|
|
one_pass: bool | None = None,
|
|
) -> NormalizationResults:
|
|
"""Normalize all unnormalized sounds."""
|
|
logger.info("Starting normalization of all sounds")
|
|
|
|
results: NormalizationResults = {
|
|
"processed": 0,
|
|
"normalized": 0,
|
|
"skipped": 0,
|
|
"errors": 0,
|
|
"files": [],
|
|
}
|
|
|
|
# Get sounds to normalize
|
|
if force:
|
|
# Get all sounds if forcing
|
|
sounds = []
|
|
for sound_type in self.type_directories.keys():
|
|
type_sounds = await self.sound_repo.get_by_type(sound_type)
|
|
sounds.extend(type_sounds)
|
|
else:
|
|
# Get only unnormalized sounds
|
|
sounds = await self.sound_repo.get_unnormalized_sounds()
|
|
|
|
logger.info("Found %d sounds to process", len(sounds))
|
|
|
|
# Capture all sound data upfront to avoid session detachment issues
|
|
sound_data_list = []
|
|
for sound in sounds:
|
|
sound_data_list.append(
|
|
{
|
|
"id": sound.id,
|
|
"filename": sound.filename,
|
|
"type": sound.type,
|
|
"is_normalized": sound.is_normalized,
|
|
"name": sound.name,
|
|
}
|
|
)
|
|
|
|
# Process each sound using captured data
|
|
for i, sound in enumerate(sounds):
|
|
results["processed"] += 1
|
|
|
|
# Use captured data to avoid detached instance issues
|
|
sound_data = sound_data_list[i]
|
|
sound_id = sound_data["id"]
|
|
sound_filename = sound_data["filename"]
|
|
|
|
try:
|
|
normalization_info = await self.normalize_sound(
|
|
sound,
|
|
force=force,
|
|
one_pass=one_pass,
|
|
sound_data=sound_data,
|
|
)
|
|
|
|
results["files"].append(normalization_info)
|
|
|
|
if normalization_info["status"] == "normalized":
|
|
results["normalized"] += 1
|
|
elif normalization_info["status"] == "skipped":
|
|
results["skipped"] += 1
|
|
elif normalization_info["status"] == "error":
|
|
results["errors"] += 1
|
|
|
|
except Exception as e:
|
|
logger.exception(
|
|
"Unexpected error processing sound %s",
|
|
sound_filename,
|
|
)
|
|
results["errors"] += 1
|
|
results["files"].append(
|
|
{
|
|
"filename": sound_filename,
|
|
"status": "error",
|
|
"reason": None,
|
|
"original_path": None,
|
|
"normalized_path": None,
|
|
"normalized_filename": None,
|
|
"normalized_duration": None,
|
|
"normalized_size": None,
|
|
"normalized_hash": None,
|
|
"id": sound_id,
|
|
"error": str(e),
|
|
}
|
|
)
|
|
|
|
logger.info("Normalization completed: %s", results)
|
|
return results
|
|
|
|
async def normalize_sounds_by_type(
|
|
self,
|
|
sound_type: str,
|
|
force: bool = False,
|
|
one_pass: bool | None = None,
|
|
) -> NormalizationResults:
|
|
"""Normalize all sounds of a specific type."""
|
|
logger.info("Starting normalization of %s sounds", sound_type)
|
|
|
|
results: NormalizationResults = {
|
|
"processed": 0,
|
|
"normalized": 0,
|
|
"skipped": 0,
|
|
"errors": 0,
|
|
"files": [],
|
|
}
|
|
|
|
# Get sounds to normalize
|
|
if force:
|
|
sounds = await self.sound_repo.get_by_type(sound_type)
|
|
else:
|
|
sounds = await self.sound_repo.get_unnormalized_sounds_by_type(sound_type)
|
|
|
|
logger.info("Found %d %s sounds to process", len(sounds), sound_type)
|
|
|
|
# Capture all sound data upfront to avoid session detachment issues
|
|
sound_data_list = []
|
|
for sound in sounds:
|
|
sound_data_list.append(
|
|
{
|
|
"id": sound.id,
|
|
"filename": sound.filename,
|
|
"type": sound.type,
|
|
"is_normalized": sound.is_normalized,
|
|
"name": sound.name,
|
|
}
|
|
)
|
|
|
|
# Process each sound using captured data
|
|
for i, sound in enumerate(sounds):
|
|
results["processed"] += 1
|
|
|
|
# Use captured data to avoid detached instance issues
|
|
sound_data = sound_data_list[i]
|
|
sound_id = sound_data["id"]
|
|
sound_filename = sound_data["filename"]
|
|
|
|
try:
|
|
normalization_info = await self.normalize_sound(
|
|
sound,
|
|
force=force,
|
|
one_pass=one_pass,
|
|
sound_data=sound_data,
|
|
)
|
|
|
|
results["files"].append(normalization_info)
|
|
|
|
if normalization_info["status"] == "normalized":
|
|
results["normalized"] += 1
|
|
elif normalization_info["status"] == "skipped":
|
|
results["skipped"] += 1
|
|
elif normalization_info["status"] == "error":
|
|
results["errors"] += 1
|
|
|
|
except Exception as e:
|
|
logger.exception(
|
|
"Unexpected error processing sound %s",
|
|
sound_filename,
|
|
)
|
|
results["errors"] += 1
|
|
results["files"].append(
|
|
{
|
|
"filename": sound_filename,
|
|
"status": "error",
|
|
"reason": None,
|
|
"original_path": None,
|
|
"normalized_path": None,
|
|
"normalized_filename": None,
|
|
"normalized_duration": None,
|
|
"normalized_size": None,
|
|
"normalized_hash": None,
|
|
"id": sound_id,
|
|
"error": str(e),
|
|
}
|
|
)
|
|
|
|
logger.info("Type normalization completed: %s", results)
|
|
return results
|