inference.py

from TTS.api import TTS
import torch 
import json


# Path to your local config and checkpoint
config_path = "./config.json"
checkpoint_path = "./pytorch_model.pth"

# Load model from checkpoint
gpu = torch.cuda.is_available()
tts = TTS(config_path=config_path, model_path=checkpoint_path, progress_bar=False, gpu=gpu)

# model_id = "bangla-speech-processing/bangla_tts_female"  # local model path or Hugging Face ID

# tts = TTS(model_name=model_id, progress_bar=False, gpu=False)

# Detect if GPU is available and initialize TTS
# gpu = torch.cuda.is_available()
# tts = TTS(model_id, progress_bar=False, gpu=gpu)

def run_tts(text, output_path="output.wav"):
    """Convert text to speech and save as file"""
    tts.tts_to_file(text=text, file_path=output_path)
    return output_path


# def text_to_speech(text):
#     output_path = "output.wav"
#     tts.tts_to_file(text=text, file_path=output_path)
#     return output_path

# Gradio app
# demo = gr.Interface(
#     fn=text_to_speech,
#     inputs="text",
#     outputs="audio",
#     title="Bangla Text to Speech",
#     description="Enter Bangla text and get speech output"
# )

# if __name__ == "__main__":
#     demo.launch()




# tts --model_path bangla_tts_female/pytorch_model.pth \
#      --config_path bangla_tts_female/config.json \
#      --text "আমি বাংলাদেশ থেকে এসেছি।" \
#      --out_path baseline.wav
     
# from IPython.display import Audio
# Audio("baseline.wav")