Upload inference.py with huggingface_hub

inference.py

@@ -0,0 +1,61 @@
+"""
+Thorsten-Voice/Kokoro - German TTS inference example.
+
+Downloads model, config and voicepack from the Hugging Face Hub
+(no local files required) and synthesizes German speech in Thorsten's voice.
+
+Usage:
+    python inference.py "Hallo, hier spricht Thorsten." output.wav
+"""
+import sys
+
+import numpy as np
+import soundfile as sf
+import torch
+from huggingface_hub import hf_hub_download
+
+REPO_ID = "Thorsten-Voice/Kokoro"
+BASE_REPO_ID = "hexgrad/Kokoro-82M"  # architecture reference only, no download from here
+SAMPLE_RATE = 24000
+
+
+def load_pipeline(device: str | None = None):
+    from kokoro import KModel, KPipeline
+
+    device = device or ("cuda" if torch.cuda.is_available() else "cpu")
+
+    config_path = hf_hub_download(repo_id=REPO_ID, filename="config.json")
+    model_path = hf_hub_download(repo_id=REPO_ID, filename="model.pth")
+    voice_path = hf_hub_download(repo_id=REPO_ID, filename="voices/thorsten.pt")
+
+    kmodel = KModel(repo_id=BASE_REPO_ID, config=config_path, model=model_path)
+    kmodel = kmodel.to(device).eval()
+
+    pipeline = KPipeline(lang_code="d", repo_id=BASE_REPO_ID, model=kmodel)
+    voice = torch.load(voice_path, map_location="cpu", weights_only=True)
+
+    return pipeline, voice, device
+
+
+def synthesize(text: str, output_path: str = "output.wav", speed: float = 1.0) -> None:
+    pipeline, voice, device = load_pipeline()
+    print(f"Using device: {device}")
+
+    audio_chunks = []
+    for _, phonemes, audio in pipeline(text, voice=voice, speed=speed):
+        print(f"Phonemes: {phonemes}")
+        audio_chunks.append(audio)
+
+    if not audio_chunks:
+        print("WARNING: no audio was generated")
+        return
+
+    combined = np.concatenate(audio_chunks)
+    sf.write(output_path, combined, SAMPLE_RATE)
+    print(f"Saved: {output_path} ({len(combined) / SAMPLE_RATE:.1f}s)")
+
+
+if __name__ == "__main__":
+    input_text = sys.argv[1] if len(sys.argv) > 1 else "Hallo, hier spricht Thorsten."
+    output_file = sys.argv[2] if len(sys.argv) > 2 else "output.wav"
+    synthesize(input_text, output_file)