app.py

import gradio as gr
import torch
import librosa
from transformers import Wav2Vec2Processor, AutoModelForCTC
import zipfile
import os
import firebase_admin
from firebase_admin import credentials, firestore
from datetime import datetime
import json
import tempfile

# Initialize Firebase
firebase_config = json.loads(os.environ.get('firebase_creds'))
cred = credentials.Certificate(firebase_config)  # Your Firebase JSON key file
firebase_admin.initialize_app(cred)
db = firestore.client()

# Load the ASR model and processor
MODEL_NAME = "eleferrand/xlsr53_Amis"
processor = Wav2Vec2Processor.from_pretrained(MODEL_NAME)
model = AutoModelForCTC.from_pretrained(MODEL_NAME)

def transcribe(audio_file):
    try:
        audio, rate = librosa.load(audio_file, sr=16000)
        input_values = processor(audio, sampling_rate=16000, return_tensors="pt").input_values

        with torch.no_grad():
            logits = model(input_values).logits
        predicted_ids = torch.argmax(logits, dim=-1)
        transcription = processor.batch_decode(predicted_ids)[0]
        return transcription.replace("[UNK]", "")
    except Exception as e:
        return f"处理文件错误: {e}"

def transcribe_both(audio_file):
    start_time = datetime.now()
    transcription = transcribe(audio_file)
    processing_time = (datetime.now() - start_time).total_seconds()
    return transcription, transcription, processing_time

def store_correction(original_transcription, corrected_transcription, audio_file, processing_time, age, native_speaker):
    try:
        audio_metadata = {}
        if audio_file and os.path.exists(audio_file):
            audio, sr = librosa.load(audio_file, sr=16000)
            duration = librosa.get_duration(y=audio, sr=sr)
            file_size = os.path.getsize(audio_file)
            audio_metadata = {'duration': duration, 'file_size': file_size}
        combined_data = {
            'original_text': original_transcription,
            'corrected_text': corrected_transcription,
            'timestamp': datetime.now().isoformat(),
            'processing_time': processing_time,
            'audio_metadata': audio_metadata,
            'audio_url': None,
            'model_name': MODEL_NAME,
            'user_info': {
                'native_amis_speaker': native_speaker,
                'age': age
            }
        }
        db.collection('transcriptions').add(combined_data)
        return "校正保存成功! (Correction saved successfully!)"
    except Exception as e:
        return f"保存失败: {e} (Error saving correction: {e})"

def prepare_download(audio_file, original_transcription, corrected_transcription):
    if audio_file is None:
        return None

    tmp_zip = tempfile.NamedTemporaryFile(delete=False, suffix=".zip")
    tmp_zip.close()
    with zipfile.ZipFile(tmp_zip.name, "w") as zf:
        if os.path.exists(audio_file):
            zf.write(audio_file, arcname="audio.wav")
        
        orig_txt = "original_transcription.txt"
        with open(orig_txt, "w", encoding="utf-8") as f:
            f.write(original_transcription)
        zf.write(orig_txt, arcname="original_transcription.txt")
        os.remove(orig_txt)

        corr_txt = "corrected_transcription.txt"
        with open(corr_txt, "w", encoding="utf-8") as f:
            f.write(corrected_transcription)
        zf.write(corr_txt, arcname="corrected_transcription.txt")
        os.remove(corr_txt)
    return tmp_zip.name

# 界面设计
with gr.Blocks(css="""
    .container { 
        max-width: 800px; 
        margin: auto; 
        padding: 20px; 
        font-family: Arial, sans-serif;
    }
    .header { 
        text-align: center; 
        margin-bottom: 30px; 
    }
    .section { 
        margin-bottom: 30px; 
        padding: 15px; 
        border: 1px solid #ddd; 
        border-radius: 8px; 
        background-color: #f9f9f9;
    }
    .section h3 {
        margin-top: 0;
        margin-bottom: 15px;
        text-align: center;
    }
    .button-row {
        display: flex;
        justify-content: center;
        gap: 10px;
        flex-wrap: wrap;
    }
    @media (max-width: 600px) {
      .gradio-row { 
          flex-direction: column; 
      }
    }
""") as demo:
    with gr.Column(elem_classes="container"):
        gr.Markdown("<h1 class='header'>阿美語轉錄與修正系統 (ASR Correction System)</h1>")
        
        with gr.Column(elem_classes="section"):
            gr.Markdown("### 步驟 1：音訊上傳與轉錄(Audio Upload & Transcription)")
            with gr.Row():
                audio_input = gr.Audio(
                    sources=["upload", "microphone"],
                    type="filepath",
                    label="音訊輸入 (Audio Input)"
                )
                transcribe_button = gr.Button("轉錄音訊 (Transcribe Audio)", variant="primary")
            proc_time_state = gr.State()
        
        with gr.Column(elem_classes="section"):
            gr.Markdown("### 步驟 2：審閱與編輯轉錄 (Review & Edit Transcription)")
            with gr.Row():
                original_text = gr.Textbox(
                    label="原始轉錄 (Original Transcription)",
                    interactive=False,
                    lines=5,
                    placeholder="謄本將在此出現... (Transcription will appear here...)"
                )
                corrected_text = gr.Textbox(
                    label="更正轉錄 (Corrected Transcription)",
                    interactive=True,
                    lines=5,
                    placeholder="在此編輯轉錄... (Edit transcription here...)"
                )
        
        with gr.Column(elem_classes="section"):
            gr.Markdown("### 步驟 3：使用者資訊 (User Information)")
            with gr.Row():
                age_input = gr.Slider(
                    minimum=0,
                    maximum=100,
                    step=1,
                    label="年齡 (Age)",
                    value=25
                )
                native_speaker_input = gr.Checkbox(
                    label="以阿美語為母語？ (Native Amis Speaker?)",
                    value=True
                )
        
        with gr.Column(elem_classes="section"):
            gr.Markdown("### 步驟 4：儲存與下載 (Save & Download)")
            with gr.Row(elem_classes="button-row"):
                save_button = gr.Button("儲存更正 (Save Correction)", variant="primary")
                save_status = gr.Textbox(
                    label="儲存狀態 (Save Status)",
                    interactive=False,
                    placeholder="狀態訊息會出現在這裡... (Status messages will appear here...)"
                )
            with gr.Row(elem_classes="button-row"):
                download_button = gr.Button("下載 ZIP 檔案 (Download ZIP)")
                download_output = gr.File(label="下載 ZIP 檔案  (Download ZIP)")

        transcribe_button.click(
            fn=transcribe_both,
            inputs=audio_input,
            outputs=[original_text, corrected_text, proc_time_state]
        )
        
        save_button.click(
            fn=store_correction,
            inputs=[original_text, corrected_text, audio_input, proc_time_state, age_input, native_speaker_input],
            outputs=save_status
        )
        
        download_button.click(
            fn=prepare_download,
            inputs=[audio_input, original_text, corrected_text],
            outputs=download_output
        )

demo.launch(share=True)