"""apptor-tiny-tts: German text -> WAV on the CPU. python speak.py "Hallo, wie geht es dir?" --voice F2 --out hallo.wav [--vocos] [--speed 1.0] Needs espeak-ng, onnxruntime, numpy, soundfile, num2words; --vocos also torch and vocos. """ import argparse import json import re from pathlib import Path import numpy as np import onnxruntime as ort import soundfile as sf from tinyvoca.text import phonemize HERE = Path(__file__).resolve().parent def main(): ap = argparse.ArgumentParser() ap.add_argument("text") ap.add_argument("--voice", default="F2", choices=["F2", "M2"]) ap.add_argument("--out", default="out.wav") ap.add_argument("--speed", type=float, default=1.0) ap.add_argument("--vocos", action="store_true", help="second stage: fine-tuned Vocos (cleaner sound)") ap.add_argument("--threads", type=int, default=4) a = ap.parse_args() meta = json.loads((HERE / "apptor-tiny-tts-de.json").read_text(encoding="utf-8")) opts = ort.SessionOptions() opts.intra_op_num_threads = a.threads session = ort.InferenceSession(str(HERE / "apptor-tiny-tts-de.onnx"), opts, providers=["CPUExecutionProvider"]) voice = meta["voices"].index(a.voice) pieces = [] for sentence in re.split(r"(?<=[.!?…])\s+", a.text.strip()): # one sentence per call (<= 512 tokens) if not sentence: continue p = phonemize(sentence) audio, _ = session.run(None, { "tokens": np.array([p.ids], np.int64), "stress": np.array([p.stress], np.int64), "voice": np.array([voice], np.int64), "speed": np.array([a.speed], np.float32), "pitch_shift": np.array([0.0], np.float32), "pitch_range": np.array([1.0], np.float32)}) pieces += [audio[0].astype(np.float32), np.zeros(int(0.12 * meta["sample_rate"]), np.float32)] audio = np.concatenate(pieces) if a.vocos: import torch from vocos import Vocos torch.set_num_threads(a.threads) voc = Vocos.from_pretrained("charactr/vocos-mel-24khz").eval() voc.load_state_dict(torch.load(HERE / "apptor-tiny-tts-vocos-de.pt", map_location="cpu", weights_only=True)) with torch.inference_mode(): x = torch.from_numpy(audio)[None] audio = voc.decode(voc.feature_extractor(x))[0, :len(audio)].numpy() peak = float(np.abs(audio).max()) if peak > 0.95: audio = audio * (0.95 / peak) sf.write(a.out, audio, meta["sample_rate"]) print(f"wrote {a.out} ({len(audio) / meta['sample_rate']:.1f} s)") if __name__ == "__main__": main()