app.py

import gradio as gr
from transformers import MarianMTModel, MarianTokenizer
import speech_recognition as sr
from gtts import gTTS
import pyttsx3
import tempfile
import re
# import os

# os.environ["no_proxy"]="localhost,127.0.0.1,::1"

def clean_text(text):
    """Clean text to handle punctuations and special characters."""
    return re.sub(r'[^\w\s,.!?;:]', '', text)

# Load translation models
tokenizer_en2zh = MarianTokenizer.from_pretrained("Helsinki-NLP/opus-mt-en-zh")
model_en2zh = MarianMTModel.from_pretrained("Helsinki-NLP/opus-mt-en-zh")

tokenizer_zh2en = MarianTokenizer.from_pretrained("Helsinki-NLP/opus-mt-zh-en")
model_zh2en = MarianMTModel.from_pretrained("Helsinki-NLP/opus-mt-zh-en")

def translate_text(input_text, target_language):
    cleaned_text = clean_text(input_text)
    if target_language == "en":
        tokenizer, model = tokenizer_zh2en, model_zh2en
    else:
        tokenizer, model = tokenizer_en2zh, model_en2zh
    
    inputs = tokenizer(cleaned_text, return_tensors="pt", padding=True, truncation=True, max_length=512)
    translated_tokens = model.generate(**inputs)
    translated_text = tokenizer.decode(translated_tokens[0], skip_special_tokens=True)
    return translated_text

def speech_to_text(audio_file):
    recognizer = sr.Recognizer()
    with sr.AudioFile(audio_file) as source:
        audio = recognizer.record(source)
    return recognizer.recognize_google(audio)

def synthesize_speech(text, language, method='gtts'):
    if method == 'gtts':
        tts = gTTS(text=text, lang=language)
        temp_audio = tempfile.NamedTemporaryFile(delete=False, suffix=".mp3")
        tts.save(temp_audio.name)
    elif method == 'pyttsx3':
        engine = pyttsx3.init()
        engine.save_to_file(text, temp_audio.name)
        engine.runAndWait()
    return temp_audio.name

def process_input(input_text=None, audio_file=None, method='gtts'):
    recognized_text = ""
    if audio_file:
        recognized_text = speech_to_text(audio_file)
    elif input_text:
        recognized_text = input_text
    else:
        return "No input provided", "", None
    
    detected_lang = "zh" if re.search(r'[\u4e00-\u9fff]', recognized_text) else "en"
    target_language = "zh" if detected_lang == "en" else "en"
    
    translated_text = translate_text(recognized_text, target_language)
    audio_path = synthesize_speech(translated_text, target_language, method=method)
    
    return recognized_text, translated_text, audio_path

# Gradio interface
# Gradio Interface
with gr.Blocks() as demo:
    gr.Markdown("## 中英文互译及语音播报")
    
    inputs = [
        gr.Textbox(label="Enter Text"),
        gr.Audio(sources=["upload","microphone"], type="filepath", label="Or speak into the microphone")
    ]
    
    translate_button = gr.Button("翻译并播报")
    
    outputs = [
        gr.Textbox(label="Recognized Text"),
        gr.Textbox(label="翻译结果"),
        gr.Audio(label="语音播报")
    ]

    translate_button.click(fn=process_input, inputs=inputs, outputs=outputs)

demo.launch(share=True)