Create app.py

app.py

@@ -0,0 +1,26 @@
+import gradio as gr
+import torch
+from transformers import AutoTokenizer, VitsModel
+import numpy as np
+
+# Load model iyo tokenizer
+model_name = "jellecali8/somali_tts_model"  # beddel hadii magaca repo-gaagu kala duwanaado
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = VitsModel.from_pretrained(model_name).to("cuda" if torch.cuda.is_available() else "cpu")
+
+def tts(text):
+    inputs = tokenizer(text, return_tensors="pt")
+    inputs = {k: v.to(model.device) for k, v in inputs.items()}
+    with torch.no_grad():
+        outputs = model(**inputs)
+    # Halkan waa inaad ku beddeshaa sida output audio laga helayo, waayo VitsModel ma soo saaro waveform toos ah
+    # Haddii aad haysato vocoder model, ku socodsii spectrogram outputs-ka halkaas
+    # Tusaalaha hoose waa placeholder:
+    audio = np.zeros(16000)  # 1 second of silence (sample rate 16k)
+    return (16000, audio)
+
+demo = gr.Interface(fn=tts,
+                    inputs=gr.Textbox(lines=2, placeholder="Ku qor qoraalka Somali..."),
+                    outputs=gr.Audio(type="numpy"))
+
+demo.launch()