Create app.py

app.py

@@ -0,0 +1,75 @@
+import os
+import tempfile
+import requests
+import torch
+import torchaudio as ta
+import gradio as gr
+
+from chatterbox.tts import ChatterboxTTS
+from huggingface_hub import hf_hub_download
+from safetensors.torch import load_file
+
+MODEL_REPO = "grandhigh/Chatterbox-TTS-Indonesian"
+CHECKPOINT_FILENAME = "t3_cfg.safetensors"
+DEVICE = "cpu"
+
+print("Loading model...")
+model = ChatterboxTTS.from_pretrained(device=DEVICE)
+
+checkpoint_path = hf_hub_download(repo_id=MODEL_REPO, filename=CHECKPOINT_FILENAME)
+t3_state = load_file(checkpoint_path, device="cpu")
+model.t3.load_state_dict(t3_state)
+
+model = model.to(DEVICE)
+model.eval()
+print("Model loaded.")
+
+def _download_audio_from_url(url: str) -> str:
+    r = requests.get(url, timeout=60)
+    r.raise_for_status()
+    tmp_wav = tempfile.NamedTemporaryFile(delete=False, suffix=".wav")
+    tmp_wav.write(r.content)
+    tmp_wav.close()
+    return tmp_wav.name
+
+def clone_voice(text: str, audio_file, audio_url: str):
+    if not text or not text.strip():
+        raise gr.Error("Text prompt tidak boleh kosong.")
+
+    prompt_path = None
+    if audio_file is not None:
+        prompt_path = audio_file
+    elif audio_url and audio_url.strip():
+        prompt_path = _download_audio_from_url(audio_url.strip())
+
+    if prompt_path is None:
+        raise gr.Error("Upload WAV atau isi audio_url.")
+
+    with torch.no_grad():
+        wav = model.generate(text.strip(), audio_prompt_path=prompt_path)
+
+    if wav.dim() == 1:
+        wav = wav.unsqueeze(0)
+
+    out_file = tempfile.NamedTemporaryFile(delete=False, suffix=".wav").name
+    ta.save(out_file, wav.cpu(), model.sr)
+    return out_file
+
+with gr.Blocks(title="Chatterbox Indonesian Voice Cloning API") as demo:
+    gr.Markdown("## Chatterbox-TTS Indonesian (Voice Cloning, CPU)")
+    text_in = gr.Textbox(label="Text Prompt", lines=4)
+    wav_in = gr.Audio(label="Upload WAV Prompt", type="filepath")
+    url_in = gr.Textbox(label="Audio URL (opsional)")
+    btn = gr.Button("Generate")
+    out_audio = gr.Audio(label="Hasil Audio", type="filepath")
+
+    btn.click(
+        fn=clone_voice,
+        inputs=[text_in, wav_in, url_in],
+        outputs=[out_audio],
+        api_name="clone_voice"
+    )
+
+if __name__ == "__main__":
+    port = int(os.getenv("PORT", os.getenv("GRADIO_SERVER_PORT", 7860)))
+    demo.launch(server_name="0.0.0.0", server_port=port)