app.py

import streamlit as st
import numpy as np
import torch
from transformers import pipeline
import librosa

# Load the pipelines
asr_pipe = pipeline("automatic-speech-recognition", model="alvanlii/whisper-small-cantonese")
translation_pipe = pipeline("translation", model="raptorkwok/cantonese-chinese-translation")
tts_pipe = pipeline("text-to-speech", model="myshell-ai/MeloTTS-Chinese")

# Streamlit UI
st.title("Cantonese to Chinese Translator")
st.write("Upload your Cantonese audio file (WAV format) below.")

# File upload
uploaded_file = st.file_uploader("Choose a WAV file", type="wav")

if uploaded_file is not None:
    # Load the audio file
    audio, sr = librosa.load(uploaded_file, sr=16000)

    # Recognize Cantonese speech
    audio_input = torch.tensor(audio)
    result = asr_pipe(audio_input)
    cantonese_text = result['text']
    st.write(f"Cantonese Text: {cantonese_text}")

    # Translate Cantonese to Chinese
    chinese_text = translation_pipe(cantonese_text)[0]['translation_text']
    st.write(f"Chinese Text: {chinese_text}")

    # Convert Chinese text to speech
    tts_output = tts_pipe(chinese_text)

    # Play back the Chinese output
    st.audio(tts_output['audio'], format='audio/wav')