Create app.py

app.py

@@ -0,0 +1,86 @@
+import os
+import torch
+import gradio as gr
+import soundfile as sf
+from transformers import AutoProcessor, VitsModel
+
+
+HF_TOKEN = os.getenv("HF_TOKEN")
+DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
+
+TTS_MODELS = {
+    "yoruba": "facebook/mms-tts-yor",
+    "hausa": "facebook/mms-tts-hau",
+}
+
+
+tts_engines = {}
+
+for lang, model_id in TTS_MODELS.items():
+    print(f"Loading TTS model for {lang}...")
+
+    processor = AutoProcessor.from_pretrained(
+        model_id,
+        token=HF_TOKEN
+    )
+
+    model = VitsModel.from_pretrained(
+        model_id,
+        token=HF_TOKEN
+    ).to(DEVICE)
+
+    model.eval()
+
+    tts_engines[lang] = {
+        "processor": processor,
+        "model": model
+    }
+
+print("All TTS models loaded successfully")
+
+
+def synthesize_speech(text, language):
+    if not text.strip():
+        return None
+
+    language = language.lower()
+    if language not in tts_engines:
+        return None
+
+    processor = tts_engines[language]["processor"]
+    model = tts_engines[language]["model"]
+
+    inputs = processor(
+        text=text,
+        return_tensors="pt"
+    ).to(DEVICE)
+
+    with torch.no_grad():
+        output = model(**inputs)
+
+    audio = output.waveform.squeeze().cpu().numpy()
+
+    output_path = "tts_output.wav"
+    sf.write(output_path, audio, 16000)
+
+    return output_path
+
+
+demo = gr.Interface(
+    fn=synthesize_speech,
+    inputs=[
+        gr.Textbox(label="Text"),
+        gr.Dropdown(
+            choices=["yoruba", "hausa"],
+            label="Language"
+        )
+    ],
+    outputs=gr.Audio(type="filepath", label="Generated Speech"),
+    title="HealthAtlas Nigerian TTS Service",
+    description="Text → Speech (Yoruba & Hausa)",
+    allow_flagging="never"
+)
+
+
+if __name__ == "__main__":
+    demo.launch()