Fabio
c36510c21e
Some checks failed
Linux Build / Linux cpu (push) Has been cancelled
Linux Build / Linux cpu-clang (push) Has been cancelled
Linux Build / Linux vulkan (push) Has been cancelled
mac-build / macOS CPU (push) Has been cancelled
Nix Build / Nix (cpu) (push) Has been cancelled
Nix Build / Nix (vulkan) (push) Has been cancelled
windows-build / Windows CPU (push) Has been cancelled
74 lines
2.5 KiB
Python
74 lines
2.5 KiB
Python
#!/usr/bin/env python3
|
|
"""Run an official Piper ONNX voice with the matching eSpeak frontend."""
|
|
|
|
from __future__ import annotations
|
|
|
|
import argparse
|
|
import json
|
|
import subprocess
|
|
import unicodedata
|
|
import wave
|
|
from pathlib import Path
|
|
|
|
import numpy as np
|
|
import onnxruntime as ort
|
|
|
|
|
|
def main() -> None:
|
|
parser = argparse.ArgumentParser()
|
|
parser.add_argument("--onnx", type=Path, required=True)
|
|
parser.add_argument("--config", type=Path, required=True)
|
|
parser.add_argument("--espeak", type=Path, required=True)
|
|
parser.add_argument("--espeak-data-parent", type=Path, required=True)
|
|
parser.add_argument("--text", required=True)
|
|
parser.add_argument("--output", type=Path, required=True)
|
|
parser.add_argument("--noise-scale", type=float, default=0.667)
|
|
parser.add_argument("--length-scale", type=float, default=1.0)
|
|
parser.add_argument("--noise-w", type=float, default=0.8)
|
|
args = parser.parse_args()
|
|
|
|
config = json.loads(args.config.read_text(encoding="utf-8"))
|
|
phonemes = subprocess.check_output(
|
|
[
|
|
str(args.espeak),
|
|
f"--path={args.espeak_data_parent}",
|
|
f"-v{config['espeak']['voice']}",
|
|
"-q",
|
|
"--ipa=3",
|
|
args.text,
|
|
],
|
|
text=True,
|
|
).strip()
|
|
phonemes = unicodedata.normalize("NFD", phonemes).replace("\u200d", "").replace("\u0361", "")
|
|
ids = [1, 0]
|
|
mapping = config["phoneme_id_map"]
|
|
for symbol in phonemes:
|
|
if symbol in mapping:
|
|
ids.extend((mapping[symbol][0], 0))
|
|
ids.append(2)
|
|
|
|
session = ort.InferenceSession(str(args.onnx), providers=["CPUExecutionProvider"])
|
|
tokens = np.asarray([ids], dtype=np.int64)
|
|
audio = session.run(
|
|
None,
|
|
{
|
|
"input": tokens,
|
|
"input_lengths": np.asarray([tokens.shape[1]], dtype=np.int64),
|
|
"scales": np.asarray(
|
|
[args.noise_scale, args.length_scale, args.noise_w], dtype=np.float32
|
|
),
|
|
},
|
|
)[0].reshape(-1)
|
|
pcm = np.clip(audio, -1.0, 1.0)
|
|
pcm = np.round(pcm * 32767.0).astype("<i2")
|
|
args.output.parent.mkdir(parents=True, exist_ok=True)
|
|
with wave.open(str(args.output), "wb") as output:
|
|
output.setnchannels(1)
|
|
output.setsampwidth(2)
|
|
output.setframerate(config["audio"]["sample_rate"])
|
|
output.writeframes(pcm.tobytes())
|
|
print(json.dumps({"phonemes": phonemes, "token_ids": ids, "samples": len(audio)}))
|
|
|
|
|
|
if __name__ == "__main__":
|
|
main()
|