Upload narrator.py with huggingface_hub
Browse files- narrator.py +132 -0
narrator.py
ADDED
|
@@ -0,0 +1,132 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/usr/bin/env python3
|
| 2 |
+
"""
|
| 3 |
+
narrator.py
|
| 4 |
+
-----------
|
| 5 |
+
Turnkey documentary voiceover generator for Kokoro Grounded Documentary.
|
| 6 |
+
Can be imported directly into any Python project or run via CLI.
|
| 7 |
+
"""
|
| 8 |
+
|
| 9 |
+
import os
|
| 10 |
+
import subprocess
|
| 11 |
+
from pathlib import Path
|
| 12 |
+
from typing import Optional, Union
|
| 13 |
+
import numpy as np
|
| 14 |
+
import soundfile as sf
|
| 15 |
+
from kokoro_onnx import Kokoro
|
| 16 |
+
|
| 17 |
+
try:
|
| 18 |
+
from huggingface_hub import hf_hub_download
|
| 19 |
+
except ImportError:
|
| 20 |
+
hf_hub_download = None
|
| 21 |
+
|
| 22 |
+
DEFAULT_BROADCAST_FILTER = (
|
| 23 |
+
"highpass=f=80,"
|
| 24 |
+
"equalizer=f=160:t=q:w=1.0:g=2.8,"
|
| 25 |
+
"equalizer=f=260:t=q:w=1.2:g=2.0,"
|
| 26 |
+
"equalizer=f=450:t=q:w=2.0:g=-2.0,"
|
| 27 |
+
"equalizer=f=3200:t=q:w=2.0:g=-1.6,"
|
| 28 |
+
"equalizer=f=5800:t=q:w=2.5:g=-3.5,"
|
| 29 |
+
"equalizer=f=7200:t=q:w=2.5:g=-3.0,"
|
| 30 |
+
"lowpass=f=11000,"
|
| 31 |
+
"deesser=i=0.45:m=0.5:f=0.55:s=o,"
|
| 32 |
+
"compand=attacks=0.03:decays=0.3:points=-80/-80|-35/-25|-16/-13|0/-3,"
|
| 33 |
+
"loudnorm=I=-16:TP=-1.5:LRA=8.5"
|
| 34 |
+
)
|
| 35 |
+
|
| 36 |
+
|
| 37 |
+
class GroundedNarrator:
|
| 38 |
+
"""
|
| 39 |
+
Broadcast-grade documentary narrator using Kokoro-ONNX and the Grounded Documentary profile.
|
| 40 |
+
"""
|
| 41 |
+
|
| 42 |
+
def __init__(
|
| 43 |
+
self,
|
| 44 |
+
model_path: Optional[Union[str, Path]] = None,
|
| 45 |
+
voices_path: Optional[Union[str, Path]] = None,
|
| 46 |
+
profile_path: Optional[Union[str, Path]] = None,
|
| 47 |
+
repo_id: Optional[str] = None
|
| 48 |
+
):
|
| 49 |
+
base_dir = Path(__file__).resolve().parent
|
| 50 |
+
|
| 51 |
+
if model_path is None:
|
| 52 |
+
local_model = base_dir / "kokoro-v1.0.onnx"
|
| 53 |
+
if local_model.exists():
|
| 54 |
+
model_path = local_model
|
| 55 |
+
elif repo_id and hf_hub_download:
|
| 56 |
+
model_path = hf_hub_download(repo_id=repo_id, filename="kokoro-v1.0.onnx")
|
| 57 |
+
else:
|
| 58 |
+
raise FileNotFoundError("kokoro-v1.0.onnx not found. Provide model_path or repo_id.")
|
| 59 |
+
|
| 60 |
+
if voices_path is None:
|
| 61 |
+
local_voices = base_dir / "voices-v1.0.bin"
|
| 62 |
+
if local_voices.exists():
|
| 63 |
+
voices_path = local_voices
|
| 64 |
+
elif repo_id and hf_hub_download:
|
| 65 |
+
voices_path = hf_hub_download(repo_id=repo_id, filename="voices-v1.0.bin")
|
| 66 |
+
else:
|
| 67 |
+
raise FileNotFoundError("voices-v1.0.bin not found. Provide voices_path or repo_id.")
|
| 68 |
+
|
| 69 |
+
if profile_path is None:
|
| 70 |
+
local_prof = base_dir / "grounded_documentary.npy"
|
| 71 |
+
if local_prof.exists():
|
| 72 |
+
profile_path = local_prof
|
| 73 |
+
elif repo_id and hf_hub_download:
|
| 74 |
+
profile_path = hf_hub_download(repo_id=repo_id, filename="grounded_documentary.npy")
|
| 75 |
+
else:
|
| 76 |
+
raise FileNotFoundError("grounded_documentary.npy not found.")
|
| 77 |
+
|
| 78 |
+
self.kokoro = Kokoro(str(model_path), str(voices_path))
|
| 79 |
+
self.voice_style = np.load(str(profile_path))
|
| 80 |
+
|
| 81 |
+
def synthesize(
|
| 82 |
+
self,
|
| 83 |
+
text: str,
|
| 84 |
+
output_path: Union[str, Path],
|
| 85 |
+
speed: float = 0.92,
|
| 86 |
+
sentence_pause: float = 0.42,
|
| 87 |
+
clause_pause: float = 0.20,
|
| 88 |
+
apply_mastering: bool = True
|
| 89 |
+
) -> float:
|
| 90 |
+
"""
|
| 91 |
+
Synthesizes text to WAV/MP3 with deliberate pacing and broadcast mastering.
|
| 92 |
+
"""
|
| 93 |
+
out_p = Path(output_path)
|
| 94 |
+
out_p.parent.mkdir(parents=True, exist_ok=True)
|
| 95 |
+
wav_path = out_p.with_suffix(".wav")
|
| 96 |
+
|
| 97 |
+
audio, sr = self.kokoro.create(
|
| 98 |
+
text=text,
|
| 99 |
+
voice=self.voice_style,
|
| 100 |
+
speed=speed,
|
| 101 |
+
lang="en-gb",
|
| 102 |
+
sentence_pause=sentence_pause,
|
| 103 |
+
clause_pause=clause_pause,
|
| 104 |
+
continuous=False
|
| 105 |
+
)
|
| 106 |
+
|
| 107 |
+
sf.write(str(wav_path), audio, sr)
|
| 108 |
+
dur = len(audio) / sr
|
| 109 |
+
|
| 110 |
+
if out_p.suffix.lower() == ".mp3" or apply_mastering:
|
| 111 |
+
mp3_path = out_p.with_suffix(".mp3")
|
| 112 |
+
cmd = ["ffmpeg", "-y", "-i", str(wav_path)]
|
| 113 |
+
if apply_mastering:
|
| 114 |
+
cmd.extend(["-af", DEFAULT_BROADCAST_FILTER])
|
| 115 |
+
cmd.extend(["-c:a", "libmp3lame", "-b:a", "192k", str(mp3_path)])
|
| 116 |
+
subprocess.run(cmd, check=True, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
|
| 117 |
+
if out_p.suffix.lower() == ".mp3" and wav_path != out_p:
|
| 118 |
+
wav_path.unlink(missing_ok=True)
|
| 119 |
+
|
| 120 |
+
return dur
|
| 121 |
+
|
| 122 |
+
|
| 123 |
+
if __name__ == "__main__":
|
| 124 |
+
import argparse
|
| 125 |
+
parser = argparse.ArgumentParser(description="Generate broadcast documentary narration")
|
| 126 |
+
parser.add_argument("text", type=str, help="Text to speak")
|
| 127 |
+
parser.add_argument("-o", "--output", type=str, default="narration.mp3", help="Output path (.mp3 or .wav)")
|
| 128 |
+
args = parser.parse_args()
|
| 129 |
+
|
| 130 |
+
narrator = GroundedNarrator()
|
| 131 |
+
dur = narrator.synthesize(args.text, args.output)
|
| 132 |
+
print(f"Generated {dur:.2f}s of mastered voiceover -> {args.output}")
|