verachen/xiaozhi-esp32
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: Audio loudness normalization for assets files (#332)

Browse Source
This commit is contained in:
laride
2025-03-13 12:14:17 +08:00
committed by GitHub
parent 7eb710d404
commit ff7f396f9d
8 changed files with 112 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files

48
scripts/p3_tools/convert_audio_to_p3.py
View File

@ -5,44 +5,58 @@ import struct
import sys
import tqdm
import numpy as np
import argparse
import pyloudnorm as pyln
def encode_audio_to_opus(input_file, output_file):
def encode_audio_to_opus(input_file, output_file, target_lufs=None):
# Load audio file using librosa
audio, sample_rate = librosa.load(input_file, sr=None, mono=False, dtype=np.float32)
# Convert to mono if stereo
if audio.ndim == 2:
audio = librosa.to_mono(audio)
if target_lufs is not None:
print("Note: Automatic loudness adjustment is enabled, which may cause", file=sys.stderr)
print(" audio distortion. If the input audio has already been ", file=sys.stderr)
print(" loudness-adjusted or if the input audio is TTS audio, ", file=sys.stderr)
print(" please use the `-d` parameter to disable loudness adjustment.", file=sys.stderr)
meter = pyln.Meter(sample_rate)
current_loudness = meter.integrated_loudness(audio)
audio = pyln.normalize.loudness(audio, current_loudness, target_lufs)
print(f"Adjusted loudness: {current_loudness:.1f} LUFS -> {target_lufs} LUFS")
# Convert sample rate to 16000Hz if necessary
target_sample_rate = 16000
if sample_rate != target_sample_rate:
audio = librosa.resample(audio, orig_sr=sample_rate, target_sr=target_sample_rate)
sample_rate = target_sample_rate
# Get left channel if stereo
if audio.ndim == 2:
audio = audio[0]
# Convert audio data back to int16 after resampling
# Convert audio data back to int16 after processing
audio = (audio * 32767).astype(np.int16)
# Initialize Opus encoder
encoder = opuslib.Encoder(sample_rate, 1, opuslib.APPLICATION_AUDIO)
# Encode audio data to Opus packets
# Save encoded data to file
# Encode and save
with open(output_file, 'wb') as f:
duration = 60 # 60ms every frame
duration = 60 # 60ms per frame
frame_size = int(sample_rate * duration / 1000)
for i in tqdm.tqdm(range(0, len(audio) - frame_size, frame_size)):
frame = audio[i:i + frame_size]
opus_data = encoder.encode(frame.tobytes(), frame_size=frame_size)
# protocol format, [1u type, 1u reserved, 2u len, data]
packet = struct.pack('>BBH', 0, 0, len(opus_data)) + opus_data
f.write(packet)
# Example usage
if len(sys.argv) != 3:
print('Usage: python convert.py <input_file> <output_file>')
sys.exit(1)
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='Convert audio to Opus with loudness normalization')
parser.add_argument('input_file', help='Input audio file')
parser.add_argument('output_file', help='Output .opus file')
parser.add_argument('-l', '--lufs', type=float, default=-16.0,
help='Target loudness in LUFS (default: -16)')
parser.add_argument('-d', '--disable-loudnorm', action='store_true',
help='Disable loudness normalization')
args = parser.parse_args()
input_file = sys.argv[1]
output_file = sys.argv[2]
encode_audio_to_opus(input_file, output_file)
target_lufs = None if args.disable_loudnorm else args.lufs
encode_audio_to_opus(args.input_file, args.output_file, target_lufs)