changed6 app.py

app.py

@@ -1,36 +1,55 @@
-import json
 import os
+from pathlib import Path
 
 import numpy as np
 import gradio as gr
-from TTS.utils.synthesizer import Synthesizer
+from huggingface_hub import hf_hub_download
+from TTS.api import TTS
 
-# ---------- Paths ----------
-MODEL_PATH = "checkpoint_80000.pth"   # or "best_model_23206.pth"
-CONFIG_PATH = "config.json"
+# ---------- Config for your private model repo ----------
+REPO_ID = "DhanuakaDev/SinTts-prev-v0.1"  # private model repo
+CHECKPOINT_FILENAME = "checkpoint_80000.pth"  # change if your file name differs
+CONFIG_FILENAME = "config.json"
 
-# ---------- Load config to get sample rate ----------
-with open(CONFIG_PATH, "r", encoding="utf-8") as f:
-    cfg = json.load(f)
+# Get token from Space secret (Settings -> Variables and secrets)
+HF_TOKEN = os.environ.get("HF_TOKEN")
 
-SAMPLE_RATE = cfg.get("audio", {}).get("sample_rate", 24000)
+# ---------- Download files from private repo ----------
+# hf_hub_download returns a local path in the cache
+checkpoint_path = hf_hub_download(
+    repo_id=REPO_ID,
+    filename=CHECKPOINT_FILENAME,
+    token=HF_TOKEN,      # required for private repos
+    repo_type="model",   # explicit, though "model" is default
+)
+
+config_path = hf_hub_download(
+    repo_id=REPO_ID,
+    filename=CONFIG_FILENAME,
+    token=HF_TOKEN,
+    repo_type="model",
+)
 
-# ---------- Load Coqui TTS Synthesizer ----------
-synthesizer = Synthesizer(
-    tts_checkpoint=MODEL_PATH,
-    tts_config_path=CONFIG_PATH,
-    use_cuda=False,   # Set True only if you enable GPU in the Space
+# ---------- Load TTS model (same style as your local script) ----------
+tts = TTS(
+    model_path=str(checkpoint_path),
+    config_path=str(config_path),
+    progress_bar=False,
+    gpu=False,   # Space uses CPU; enable GPU only if you switch hardware
 )
 
+SAMPLE_RATE = tts.synthesizer.output_sample_rate
+
 # ---------- Inference function ----------
 def tts_generate(text: str):
-    if not text.strip():
+    text = text.strip()
+    if not text:
         return None
 
-    # Whatever Coqui returns (list, numpy array, etc.)
-    wav = synthesizer.tts(text)
+    # Generate audio (same call as in your local script)
+    wav = tts.tts(text)
 
-    # Convert to a 1D numpy array for Gradio
+    # Ensure numpy 1D array for Gradio
     wav = np.asarray(wav, dtype="float32").flatten()
 
     # Gradio Audio(type="numpy") expects (sample_rate, np.ndarray)
@@ -48,8 +67,8 @@ demo = gr.Interface(
         label="Generated speech",
         type="numpy",
     ),
-    title="Sinhala TTS ",
-    description="Sinhala TTS model research-stage model.",
+    title="Sinhala TTS (Coqui VITS)",
+    description="Fine-tuned Sinhala TTS model using Coqui-TTS.",
 )
 
 if __name__ == "__main__":

checkpoint_80000.pth

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:a0952ffa374cd2a2c48b9f6e7c7b917052e1502a21d19f675b2523a34d66fbe6
-size 997797878

requirements.txt

@@ -9,3 +9,4 @@ soundfile
 librosa
 numpy
 scipy
+huggingface_hub