wechat-miniprogram-translator-hf/app.py

import gradio as gr
from transformers import pipeline
import torch

# 1. Load Models (this will happen only once when the app starts)
print("Loading models...")

# ASR Pipeline
asr_pipeline = pipeline(
    "automatic-speech-recognition",
    model="openai/whisper-small",
    torch_dtype=torch.float16, # Use float16 for faster inference
    device="cpu" # Specify CPU device
)

# Translation Pipelines
translators = {
    "en-zh": pipeline("translation", model="Helsinki-NLP/opus-mt-en-zh"),
    "zh-en": pipeline("translation", model="Helsinki-NLP/opus-mt-zh-en"),
    "en-ja": pipeline("translation", model="Helsinki-NLP/opus-mt-en-ja"),
    "ja-en": pipeline("translation", model="Helsinki-NLP/opus-mt-ja-en"),
    "en-ko": pipeline("translation", model="Helsinki-NLP/opus-mt-en-ko"),
    "ko-en": pipeline("translation", model="Helsinki-NLP/opus-mt-ko-en"),
}

print("Models loaded successfully.")

# 2. Define Processing Functions

def transcribe_audio(audio_file):
    print(f"Received audio file: {audio_file}")
    if audio_file is None:
        return ""
    try:
        # The pipeline expects a file path
        text = asr_pipeline(audio_file)["text"]
        print(f"ASR result: {text}")
        return text
    except Exception as e:
        print(f"Error in ASR: {e}")
        return f"Error in ASR: {e}"

def translate_text(text, source_lang, target_lang):
    print(f"Translating '{text}' from {source_lang} to {target_lang}")
    if not text:
        return ""

    # Direct translation if possible
    if f"{source_lang}-{target_lang}" in translators:
        translator = translators[f"{source_lang}-{target_lang}"]
        translated_text = translator(text)[0]['translation_text']
    # Bridge translation via English
    elif source_lang != 'en' and target_lang != 'en':
        to_english_translator = translators[f"{source_lang}-en"]
        english_text = to_english_translator(text)[0]['translation_text']
        
        from_english_translator = translators[f"en-{target_lang}"]
        translated_text = from_english_translator(english_text)[0]['translation_text']
    else:
        return "Translation route not supported"

    print(f"Translation result: {translated_text}")
    return translated_text

# 3. Create Gradio Interface
with gr.Blocks() as demo:
    gr.Markdown("## All-in-One ASR and Translation API")
    
    with gr.Tab("ASR"):
        audio_input = gr.Audio(type="filepath", label="Upload Audio")
        asr_output = gr.Textbox(label="Transcript")
        asr_button = gr.Button("Transcribe")
        asr_button.click(transcribe_audio, inputs=audio_input, outputs=asr_output, api_name="asr")

    with gr.Tab("Translate"):
        text_input = gr.Textbox(label="Input Text")
        source_lang_input = gr.Dropdown(["en", "zh", "ja", "ko"], label="Source Language")
        target_lang_input = gr.Dropdown(["en", "zh", "ja", "ko"], label="Target Language")
        translation_output = gr.Textbox(label="Translation")
        translate_button = gr.Button("Translate")
        translate_button.click(translate_text, inputs=[text_input, source_lang_input, target_lang_input], outputs=translation_output, api_name="translate")

# 4. Launch the App
if __name__ == "__main__":
    demo.launch()