File size: 3,787 Bytes
97e3499 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 | """
Phase 8: Studio-Quality Audio Upsampling
Uses AudioSR to upsample XTTS-v2 output from 24kHz to 48kHz,
reconstructing high-frequency detail that the base model cannot produce.
The XTTS-v2 model outputs 24kHz audio. Changing the internal sample rate
would require retraining the entire model (~370M params) from scratch.
Instead, we use a neural upsampler as a post-processing step — the same
approach used in professional TTS production pipelines.
Usage:
conda activate new-arabic-tts
python scripts/upsample.py --input outputs/finetuned_model_test.wav
Output:
Upsampled WAV saved alongside the input with _48kHz suffix.
Usage for all outputs:
python scripts/upsample.py --all
"""
import os
import sys
import argparse
import time
import json
import numpy as np
import soundfile as sf
from pathlib import Path
PROJECT_ROOT = Path(__file__).resolve().parent.parent
def upsample_audiosr(input_path, output_path):
"""Upsample using AudioSR neural upsampler."""
import audiosr
print(f" Loading AudioSR model...")
model = audiosr.build_model(model_name="speech", device="cuda")
print(f" Upsampling: {input_path}")
t0 = time.time()
waveform = audiosr.super_resolution(
model,
str(input_path),
seed=42,
guidance_scale=3.5,
ddim_steps=50,
)
elapsed = time.time() - t0
# AudioSR returns [batch, channels, samples] at 48kHz
if hasattr(waveform, 'cpu'):
wav = waveform[0, 0].cpu().numpy()
else:
wav = waveform[0, 0]
target_sr = 48000
sf.write(str(output_path), wav, target_sr, subtype="PCM_24")
duration = len(wav) / target_sr
print(f" Saved: {output_path}")
print(f" Duration: {duration:.2f}s, Sample rate: {target_sr}Hz, Bit depth: 24-bit")
print(f" Processing time: {elapsed:.1f}s")
return {
"input": str(input_path),
"output": str(output_path),
"input_sr": 24000,
"output_sr": target_sr,
"bit_depth": 24,
"duration_s": round(duration, 2),
"processing_time_s": round(elapsed, 1),
}
def main():
parser = argparse.ArgumentParser(description="Audio Upsampling (24kHz → 48kHz)")
parser.add_argument("--input", type=str, help="Input WAV file to upsample")
parser.add_argument("--all", action="store_true", help="Upsample all official output files")
args = parser.parse_args()
if not args.input and not args.all:
parser.error("Provide --input <file> or --all")
print("=" * 70)
print(" Phase 8: Studio-Quality Audio Upsampling")
print(" 24kHz / 16-bit → 48kHz / 24-bit")
print("=" * 70)
results = []
if args.all:
files = [
PROJECT_ROOT / "outputs" / "base_model_test.wav",
PROJECT_ROOT / "outputs" / "finetuned_model_test.wav",
]
else:
files = [Path(args.input)]
for input_path in files:
if not input_path.exists():
print(f"\n WARNING: {input_path} not found, skipping")
continue
output_path = input_path.with_stem(input_path.stem + "_48kHz")
print(f"\n [{files.index(input_path)+1}/{len(files)}] {input_path.name}")
result = upsample_audiosr(input_path, output_path)
results.append(result)
# Save report
benchmarks_dir = PROJECT_ROOT / "docs" / "benchmarks"
benchmarks_dir.mkdir(parents=True, exist_ok=True)
report_path = benchmarks_dir / "upsampling.json"
with open(report_path, "w", encoding="utf-8") as f:
json.dump({"date": time.strftime("%Y-%m-%d"), "results": results}, f, indent=2)
print(f"\n{'='*70}")
print(f" Upsampling Complete!")
print(f" Report: {report_path}")
print(f"{'='*70}")
if __name__ == "__main__":
main()
|