Chatterbox-Multilingual-TTS / engine /audio_processor.py
flozi00's picture
Refactor audio processing to use numpy for mixing and normalization, improving performance and clarity
73f01a4
"""
Audio post-processing for phone announcements.
Handles background music mixing, normalization, and export.
"""
import io
import os
from dataclasses import dataclass
from pathlib import Path
from typing import Optional, Union
import numpy as np
from loguru import logger
@dataclass
class AudioProcessingConfig:
"""Configuration for audio post-processing."""
# Background music settings
background_music_path: Optional[str] = None
music_volume_db: float = -20.0 # Relative volume of background music
# Fade settings
fade_in_ms: int = 500
fade_out_ms: int = 500
# Padding (silence before/after speech)
padding_start_ms: int = 300
padding_end_ms: int = 300
# Output settings
normalize: bool = True
target_loudness_db: float = -16.0 # Target LUFS for normalization
output_sample_rate: int = 44100
output_format: str = "mp3"
class AudioProcessor:
"""
Post-processor for TTS audio.
Adds background music, applies fades, normalizes, and exports.
"""
# Default background music directory
ASSETS_DIR = Path(__file__).parent / "data" / "assets"
def __init__(self, config: Optional[AudioProcessingConfig] = None):
self.config = config or AudioProcessingConfig()
def process(
self,
audio: np.ndarray,
sample_rate: int,
output_path: Optional[str] = None,
**override_config,
) -> Union[bytes, str]:
"""
Process audio with background music, fades, and normalization.
Args:
audio: Input audio as numpy array
sample_rate: Sample rate of input audio
output_path: Optional path to save the output (returns bytes if None)
**override_config: Override any config settings for this call
Returns:
Path to output file if output_path is provided, otherwise MP3 bytes
"""
# Merge config overrides
config = AudioProcessingConfig(**{**self.config.__dict__, **override_config})
# Work in numpy to avoid pydub.set_frame_rate (it changes speed/pitch).
speech = self._ensure_mono_float32(audio)
speech_sr = int(sample_rate)
# Boost speech slightly for clarity (+3 dB)
speech = self._apply_gain_db(speech, 3.0)
# Normalize speech BEFORE adding music so music doesn't make speech quieter.
if config.normalize:
speech = self._normalize_numpy(speech, config.target_loudness_db)
# Add padding (silence before/after speech)
if config.padding_start_ms > 0:
pad = int(round(speech_sr * (config.padding_start_ms / 1000.0)))
if pad > 0:
speech = np.concatenate([np.zeros(pad, dtype=np.float32), speech])
if config.padding_end_ms > 0:
pad = int(round(speech_sr * (config.padding_end_ms / 1000.0)))
if pad > 0:
speech = np.concatenate([speech, np.zeros(pad, dtype=np.float32)])
mixed = speech
# Mix with background music if specified
if config.background_music_path:
mixed = self._mix_background_music_numpy(
speech=mixed,
speech_sample_rate=speech_sr,
music_path=config.background_music_path,
music_gain_db=config.music_volume_db,
)
# Apply fades to the final mix
if config.fade_in_ms > 0:
mixed = self._apply_fade_in(mixed, speech_sr, config.fade_in_ms)
if config.fade_out_ms > 0:
mixed = self._apply_fade_out(mixed, speech_sr, config.fade_out_ms)
# Resample the final audio to the requested output sample rate (real resampling)
out_sr = int(config.output_sample_rate)
if speech_sr != out_sr:
mixed = self._resample_numpy(mixed, orig_sr=speech_sr, target_sr=out_sr)
# Prevent clipping after mixing
mixed = self._peak_limit(mixed, peak=0.98)
# Export via pydub (mp3/wav/etc.)
audio_segment = self._numpy_to_audiosegment(mixed, out_sr)
if output_path:
audio_segment.export(output_path, format=config.output_format)
return output_path
buffer = io.BytesIO()
audio_segment.export(buffer, format=config.output_format)
return buffer.getvalue()
def _numpy_to_audiosegment(
self, audio: np.ndarray, sample_rate: int
) -> "AudioSegment":
"""Convert numpy array to pydub AudioSegment."""
from pydub import AudioSegment
# Ensure float32 and normalize
if audio.dtype != np.float32:
audio = audio.astype(np.float32)
# Clip and convert to int16
audio = np.clip(audio, -1.0, 1.0)
audio_int16 = (audio * 32767).astype(np.int16)
# Create AudioSegment
return AudioSegment(
data=audio_int16.tobytes(),
sample_width=2, # 16-bit
frame_rate=sample_rate,
channels=1, # Mono
)
def _resolve_music_path(self, music_path: str) -> str:
"""Resolve a preset music name/path to an existing file path."""
if not music_path:
return music_path
# Absolute path
if os.path.isabs(music_path) and os.path.exists(music_path):
return music_path
# Relative / preset name: try assets dir.
candidate = self.ASSETS_DIR / f"{music_path}.mp3"
if candidate.exists():
return str(candidate)
candidate = self.ASSETS_DIR / music_path
if candidate.exists():
return str(candidate)
return music_path
def _load_audio_file_numpy(self, path: str) -> tuple[np.ndarray, int]:
"""Load an audio file to mono float32 numpy, returning (samples, sample_rate)."""
from pydub import AudioSegment
seg = AudioSegment.from_file(path)
seg = seg.set_channels(1)
sample_rate = int(seg.frame_rate)
samples = np.array(seg.get_array_of_samples())
# Convert PCM integers to float32 in [-1, 1]
max_val = float(1 << (8 * seg.sample_width - 1))
audio = (samples.astype(np.float32) / max_val).clip(-1.0, 1.0)
return audio, sample_rate
def _resample_numpy(
self, audio: np.ndarray, orig_sr: int, target_sr: int
) -> np.ndarray:
if orig_sr == target_sr:
return audio
import librosa
return librosa.resample(audio, orig_sr=orig_sr, target_sr=target_sr)
def _mix_background_music_numpy(
self,
speech: np.ndarray,
speech_sample_rate: int,
music_path: str,
music_gain_db: float,
) -> np.ndarray:
"""Mix background music into speech (numpy domain) without changing speech tempo."""
resolved = self._resolve_music_path(music_path)
if not os.path.exists(resolved):
logger.warning(f"Background music not found: {resolved}")
return speech
try:
music, music_sr = self._load_audio_file_numpy(resolved)
if music_sr != speech_sample_rate:
music = self._resample_numpy(
music, orig_sr=music_sr, target_sr=speech_sample_rate
)
# Loop / trim to match speech length
if music.size == 0:
return speech
if music.shape[0] < speech.shape[0]:
reps = int(np.ceil(speech.shape[0] / music.shape[0]))
music = np.tile(music, reps)
music = music[: speech.shape[0]]
# Apply music gain
music = self._apply_gain_db(music, music_gain_db)
return (speech + music).astype(np.float32)
except Exception as e:
logger.error(f"Failed to add background music: {e}")
return speech
def _ensure_mono_float32(self, audio: np.ndarray) -> np.ndarray:
audio = np.asarray(audio)
if audio.ndim == 2:
# If (n, channels), downmix
audio = audio.mean(axis=1)
if audio.dtype != np.float32:
audio = audio.astype(np.float32)
return np.clip(audio, -1.0, 1.0)
def _apply_gain_db(self, audio: np.ndarray, gain_db: float) -> np.ndarray:
factor = float(10 ** (gain_db / 20.0))
return (audio * factor).astype(np.float32)
def _rms_dbfs(self, audio: np.ndarray) -> float:
rms = float(np.sqrt(np.mean(np.square(audio), dtype=np.float64)))
return float(20.0 * np.log10(rms + 1e-9))
def _normalize_numpy(self, audio: np.ndarray, target_dbfs: float) -> np.ndarray:
current = self._rms_dbfs(audio)
gain_db = float(target_dbfs - current)
return self._apply_gain_db(audio, gain_db)
def _apply_fade_in(
self, audio: np.ndarray, sample_rate: int, fade_ms: int
) -> np.ndarray:
n = int(round(sample_rate * (fade_ms / 1000.0)))
if n <= 0:
return audio
n = min(n, audio.shape[0])
ramp = np.linspace(0.0, 1.0, n, dtype=np.float32)
out = audio.copy()
out[:n] *= ramp
return out
def _apply_fade_out(
self, audio: np.ndarray, sample_rate: int, fade_ms: int
) -> np.ndarray:
n = int(round(sample_rate * (fade_ms / 1000.0)))
if n <= 0:
return audio
n = min(n, audio.shape[0])
ramp = np.linspace(1.0, 0.0, n, dtype=np.float32)
out = audio.copy()
out[-n:] *= ramp
return out
def _peak_limit(self, audio: np.ndarray, peak: float = 0.98) -> np.ndarray:
max_abs = float(np.max(np.abs(audio))) if audio.size else 0.0
if max_abs <= 0:
return audio
if max_abs <= peak:
return np.clip(audio, -1.0, 1.0)
scale = float(peak / max_abs)
return np.clip(audio * scale, -1.0, 1.0).astype(np.float32)
def list_available_music(self) -> list[str]:
"""List available background music files in the assets directory."""
logger.debug(f"Looking for music in: {self.ASSETS_DIR}")
logger.debug(f"ASSETS_DIR exists: {self.ASSETS_DIR.exists()}")
if not self.ASSETS_DIR.exists():
logger.warning(f"Assets directory not found: {self.ASSETS_DIR}")
return []
music_files = []
for ext in ["mp3", "wav", "flac", "ogg"]:
found = list(self.ASSETS_DIR.glob(f"*.{ext}"))
logger.debug(f"Found {len(found)} .{ext} files")
music_files.extend([f.stem for f in found])
result = sorted(set(music_files))
logger.info(f"Available background music: {result}")
return result