Create app.py

app.py

@@ -0,0 +1,67 @@
+import numpy as np
+import soundfile as sf
+from safetensors.numpy import load_file
+import sentencepiece as spm
+import torch
+
+# Load tokenizer
+sp = spm.SentencePieceProcessor()
+sp.load("tokenizer.model")
+
+# Load quantized model
+tensors = load_file("model.safetensors")
+
+# Dequantize weights
+weights = {}
+
+for name in list(tensors.keys()):
+
+    if name.endswith("_scale"):
+        continue
+
+    scale_name = name + "_scale"
+
+    if scale_name in tensors:
+
+        weight_i8 = tensors[name].astype(np.float32)
+        scale = tensors[scale_name].astype(np.float32)
+
+        weights[name] = weight_i8 * scale
+
+    else:
+        weights[name] = tensors[name]
+
+
+print("Model loaded successfully")
+
+
+# Dummy inference function (example structure)
+# NOTE: Pocket-TTS requires full architecture,
+# this example shows structure and audio output pipeline
+
+def generate_dummy_audio(text):
+
+    tokens = sp.encode(text)
+
+    print("Tokens:", tokens)
+
+    # Generate dummy waveform (replace with real inference)
+    duration = 3  # seconds
+    sample_rate = 24000
+
+    t = np.linspace(0, duration, int(sample_rate * duration))
+
+    audio = 0.2 * np.sin(2 * np.pi * 220 * t)
+
+    return audio, sample_rate
+
+
+# Text input
+text = "Hello Subiksha, welcome to text to speech system"
+
+audio, sr = generate_dummy_audio(text)
+
+# Save audio
+sf.write("output.wav", audio, sr)
+
+print("Speech saved as output.wav")