from TTSInferencing import TTSInferencing
from speechbrain.inference.vocoders import HIFIGAN
# import torchaudio
import  streamlit as st
import numpy as np


tts_model = TTSInferencing.from_hparams(source="./",
                                        hparams_file='./hyperparams.yaml',
                                        pymodule_file='./module_classes.py',
                                        # savedir="./",
                                        )

hifi_gan = HIFIGAN.from_hparams(source="speechbrain/tts-hifigan-ljspeech")

# text = ["Hello I am a girl", "How is your day going", "I hope you are doing well"]

# Input text
text = st.text_input("Enter your text here")

if text:
    mel_outputs = tts_model.encode_batch(text)
    waveforms = hifi_gan.decode_batch(mel_outputs)

    waveform =  waveforms[0].squeeze(1).numpy()

    # Normalize the waveform to the range [-1, 1] if necessary
    if np.max(np.abs(waveform)) > 1.0:
        waveform /= np.max(np.abs(waveform))

    # Display the audio widget to play the synthesized speech
    st.audio(waveform, format="audio/wav", sample_rate = 22050)