Update app.py

app.py

@@ -1,222 +1,222 @@
-import gradio as gr
-from google import genai
-import nemo.collections.asr as nemo_asr
-from pydub import AudioSegment
-import os
-from huggingface_hub import login
-from hazm import Normalizer
-import numpy as np
-import re
-
-# Fetch API Keys from environment variables
-HF_TOKEN = os.getenv("HF_TOKEN")
-GEMINI_API_KEY = os.getenv("GEMINI_API_KEY")
-
-if not HF_TOKEN:
-    raise ValueError("HF_TOKEN environment variable not set. Please provide a valid Hugging Face token.")
-
-if not GEMINI_API_KEY:
-    raise ValueError("GEMINI_API_KEY environment variable not set. Please provide a valid GEMINI_API_KEY.")
-
-# Authenticate with Hugging Face
-login(HF_TOKEN)
-
-# Load the NeMo ASR Model
-try:
-    asr_model = nemo_asr.models.EncDecHybridRNNTCTCBPEModel.from_pretrained(
-        model_name="faimlab/stt_fa_fastconformer_hybrid_large_dataset_v30"
-    )
-except Exception as e:
-    raise RuntimeError(f"Failed to load model: {str(e)}")
-
-# Persian text normalizer
-normalizer = Normalizer()
-
-# Function to load audio
-def load_audio(audio_path):
-    audio = AudioSegment.from_file(audio_path)
-    audio = audio.set_channels(1).set_frame_rate(16000)
-    audio_samples = np.array(audio.get_array_of_samples(), dtype=np.float32)
-    audio_samples /= np.max(np.abs(audio_samples))
-    return audio_samples, audio.frame_rate
-
-# Transcribe audio chunk
-def transcribe_chunk(audio_chunk, model):
-    transcription = model.transcribe([audio_chunk], batch_size=1, verbose=False)
-    return transcription[0].text
-
-# Transcribe the full audio file
-def transcribe_audio(file_path, model, chunk_size=30*16000):
-    waveform, _ = load_audio(file_path)
-    transcriptions = []
-    for start in range(0, len(waveform), chunk_size):
-        end = min(len(waveform), start + chunk_size)
-        transcription = transcribe_chunk(waveform[start:end], model)
-        transcriptions.append(transcription)
-
-    transcriptions = ' '.join(transcriptions)
-    transcriptions = re.sub(' +', ' ', transcriptions)
-    transcriptions = normalizer.normalize(transcriptions)
-    
-    return transcriptions
-
-# Main transcription function
-def transcribe(audio):
-    if audio is None:
-        return "Please upload an audio file.", gr.update(interactive=False), gr.update(interactive=False)
-
-    transcription = transcribe_audio(audio, asr_model)
-
-    # Enable summarize & translate buttons once transcription is done
-    return transcription, gr.update(interactive=True), gr.update(interactive=True), gr.update(interactive=True)
-
-# List of 100 languages for summarization
-languages = [
-    "English", "Persian", "French", "Spanish", "German", "Italian", "Portuguese", "Dutch", "Swedish", "Danish",
-    "Finnish", "Norwegian", "Russian", "Polish", "Turkish", "Arabic", "Hindi", "Chinese", "Japanese", "Korean",
-    "Thai", "Vietnamese", "Indonesian", "Hebrew", "Greek", "Czech", "Hungarian", "Romanian", "Bulgarian", "Serbian",
-    "Croatian", "Slovak", "Slovenian", "Ukrainian", "Lithuanian", "Latvian", "Estonian", "Macedonian", "Albanian",
-    "Basque", "Catalan", "Maltese", "Icelandic", "Georgian", "Armenian", "Belarusian", "Yiddish", "Pashto", "Urdu",
-    "Bengali", "Punjabi", "Tamil", "Telugu", "Malayalam", "Sinhala", "Burmese", "Lao", "Khmer", "Mongolian",
-    "Nepali", "Marathi", "Gujarati", "Kannada", "Odia", "Assamese", "Maithili", "Kurdish", "Azerbaijani", "Kazakh",
-    "Uzbek", "Turkmen", "Tajik", "Kyrgyz", "Uighur", "Tatar", "Haitian Creole", "Swahili", "Hausa", "Yoruba",
-    "Zulu", "Xhosa", "Amharic", "Somali", "Tigrinya", "Shona", "Igbo", "Malagasy", "Quechua", "Aymara", "Guarani",
-    "Sundanese", "Javanese", "Filipino", "Hmong", "Fijian", "Tongan", "Samoan", "Chamorro", "Hawaiian"
-]
-languages = sorted(languages)
-
-# List of available AI models
-model_selections = [
-    'gemini-2.0-flash',
-    'gemini-2.0-pro-exp-02-05',
-    'gemini-2.0-flash-lite'
-]
-
-def translate(text, target_language, model_sel):
-    client = genai.Client(api_key=GEMINI_API_KEY)
-    prompt = f"Translate the following text into {target_language}. Only reply with the translation.\n'{text}'."
-    
-    response = client.models.generate_content(
-        model=model_sel,
-        contents=[prompt]
-    )
-    
-    return response.text
-
-# Function to summarize transcribed text
-def summarize(transcript_text, word_count, model_sel, lang_sel):
-    client = genai.Client(api_key=GEMINI_API_KEY)
-    prompt = f"Summarize the following text in {word_count} words in {lang_sel}: '{transcript_text}'."
-    
-    response = client.models.generate_content(
-        model=model_sel,
-        contents=[prompt]
-    )
-    
-    return response.text
-
-def punctuate(transcript, model_sel):
-    client = genai.Client(api_key=GEMINI_API_KEY)
-    prompt = f"Restore puncutations of the transcript from an ASR model given. Maintain the original content. Only reply with the output. text: \n{transcript}"
-
-    response = client.models.generate_content(
-        model=model_sel,
-        contents=[prompt]
-    )
-    
-    return response.text
-
-# Gradio Interface
-with gr.Blocks(theme="huggingface") as demo:
-    gr.Markdown(
-        """
-        # 📝 Persian ASR, Translation & Summarization  
-        
-        Welcome to the **Persian Speech-to-Text & NLP** platform! This app allows you to upload an audio file,  
-        get an accurate **transcription**, and enhance the output with **translation**, **summarization**,  
-        and **punctuation restoration**.
-
-        ## 🎯 How It Works  
-        1️⃣ Upload an **audio file** containing Persian speech.  
-        2️⃣ Click **"Transcribe"** to generate the text output.  
-        3️⃣ Use additional features: **Translate**, **Summarize**, or **Restore Punctuation**.  
-        4️⃣ Customize settings: Select a **language**, **AI model**, and **summary length**.  
-        5️⃣ View and copy the processed text for your use!  
-
-        """
-    )
-
-    with gr.Row():
-        audio_input = gr.Audio(type="filepath", label="🎵 Upload Audio File")
-        transcript_output = gr.Textbox(label="📝 Transcription", interactive=False)
-        translation_output = gr.Textbox(label="🌍 Translation", interactive=False)
-        summarized_output = gr.Textbox(label="📖 Summarized Text", interactive=False)
-
-    transcribe_button = gr.Button("🎙️ Transcribe")
-    translate_button = gr.Button("🌐 Translate", interactive=False)
-    summarize_button = gr.Button("✂️ Summarize", interactive=False)
-    punctuate_button = gr.Button("🔤 Restore Punctuation", interactive=False)
-
-    with gr.Row():
-        word_count_input = gr.Number(value=50, label="📏 Word Count for Summary")
-        lang_selection = gr.Dropdown(choices=languages, value="English", label="🌎 Select Language")
-        model_selection = gr.Dropdown(choices=model_selections, value="gemini-2.0-flash", label="🤖 Select AI Model")
-
-    # Button Click Actions
-    transcribe_button.click(
-        transcribe, 
-        inputs=audio_input, 
-        outputs=[transcript_output, summarize_button, translate_button, punctuate_button]
-    )
-
-    translate_button.click(
-        translate, 
-        inputs=[transcript_output, lang_selection, model_selection], 
-        outputs=translation_output
-    )
-
-    summarize_button.click(
-        summarize, 
-        inputs=[transcript_output, word_count_input, model_selection, lang_selection], 
-        outputs=summarized_output
-    )
-
-    punctuate_button.click(  
-        punctuate,  
-        inputs=[transcript_output, model_selection],  
-        outputs=transcript_output  
-    )
-
-    gr.Markdown(
-        """
-        \n\n
-        ---
-
-        Powered by NVIDIA’s **NeMo Fast Conformer**, this tool is optimized for high-quality **Persian ASR (Automatic Speech Recognition)**. 
-
-        **📚 Trained on 800+ Hours of Speech Data:**  
-        - Common Voice 17 (~300 hours)  
-        - YouTube (~400 hours)  
-        - NasleMana (~90 hours)  
-        - In-house dataset (~70 hours) 
-
-        ---
-
-        ## 📜 License & Business Inquiries  
-
-        This application is licensed under **Creative Commons Attribution-NonCommercial 4.0 (CC BY-NC 4.0)**.
-        - **🛑 Non-Commercial Use Only** – Commercial use is not permitted without prior approval.  
-        - **🔗 Attribution Required** – Credit must be given to FAIM Group, Sharif University of Technology.  
-        - **❌ No Derivatives** – Modifications or adaptations of this work are not allowed.  
-
-        📜 Full License Details: [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/)  
-
-        📩 **Business Inquiries:**  
-        If you're interested in commercial applications, please contact us at:  
-        ✉️ **Email:** [[email protected]](mailto:[email protected])  
-
-        ---
-        """
-    )
-# Launch Gradio app
-demo.launch()
+import gradio as gr
+from google import genai
+import nemo.collections.asr as nemo_asr
+from pydub import AudioSegment
+import os
+from huggingface_hub import login
+from hazm import Normalizer
+import numpy as np
+import re
+
+# Fetch API Keys from environment variables
+HF_TOKEN = os.getenv("HF_TOKEN")
+GEMINI_API_KEY = os.getenv("GEMINI_API_KEY")
+
+if not HF_TOKEN:
+    raise ValueError("HF_TOKEN environment variable not set. Please provide a valid Hugging Face token.")
+
+if not GEMINI_API_KEY:
+    raise ValueError("GEMINI_API_KEY environment variable not set. Please provide a valid GEMINI_API_KEY.")
+
+# Authenticate with Hugging Face
+login(HF_TOKEN)
+
+# Load the NeMo ASR Model
+try:
+    asr_model = nemo_asr.models.EncDecHybridRNNTCTCBPEModel.from_pretrained(
+        model_name="faimlab/stt_fa_fastconformer_hybrid_large_dataset_v30"
+    )
+except Exception as e:
+    raise RuntimeError(f"Failed to load model: {str(e)}")
+
+# Persian text normalizer
+normalizer = Normalizer()
+
+# Function to load audio
+def load_audio(audio_path):
+    audio = AudioSegment.from_file(audio_path)
+    audio = audio.set_channels(1).set_frame_rate(16000)
+    audio_samples = np.array(audio.get_array_of_samples(), dtype=np.float32)
+    audio_samples /= np.max(np.abs(audio_samples))
+    return audio_samples, audio.frame_rate
+
+# Transcribe audio chunk
+def transcribe_chunk(audio_chunk, model):
+    transcription = model.transcribe([audio_chunk], batch_size=1, verbose=False)
+    return transcription[0].text
+
+# Transcribe the full audio file
+def transcribe_audio(file_path, model, chunk_size=30*16000):
+    waveform, _ = load_audio(file_path)
+    transcriptions = []
+    for start in range(0, len(waveform), chunk_size):
+        end = min(len(waveform), start + chunk_size)
+        transcription = transcribe_chunk(waveform[start:end], model)
+        transcriptions.append(transcription)
+
+    transcriptions = ' '.join(transcriptions)
+    transcriptions = re.sub(' +', ' ', transcriptions)
+    transcriptions = normalizer.normalize(transcriptions)
+    
+    return transcriptions
+
+# Main transcription function
+def transcribe(audio):
+    if audio is None:
+        return "Please upload an audio file.", gr.update(interactive=False), gr.update(interactive=False)
+
+    transcription = transcribe_audio(audio, asr_model)
+
+    # Enable summarize & translate buttons once transcription is done
+    return transcription
+
+# List of 100 languages for summarization
+languages = [
+    "English", "Persian", "French", "Spanish", "German", "Italian", "Portuguese", "Dutch", "Swedish", "Danish",
+    "Finnish", "Norwegian", "Russian", "Polish", "Turkish", "Arabic", "Hindi", "Chinese", "Japanese", "Korean",
+    "Thai", "Vietnamese", "Indonesian", "Hebrew", "Greek", "Czech", "Hungarian", "Romanian", "Bulgarian", "Serbian",
+    "Croatian", "Slovak", "Slovenian", "Ukrainian", "Lithuanian", "Latvian", "Estonian", "Macedonian", "Albanian",
+    "Basque", "Catalan", "Maltese", "Icelandic", "Georgian", "Armenian", "Belarusian", "Yiddish", "Pashto", "Urdu",
+    "Bengali", "Punjabi", "Tamil", "Telugu", "Malayalam", "Sinhala", "Burmese", "Lao", "Khmer", "Mongolian",
+    "Nepali", "Marathi", "Gujarati", "Kannada", "Odia", "Assamese", "Maithili", "Kurdish", "Azerbaijani", "Kazakh",
+    "Uzbek", "Turkmen", "Tajik", "Kyrgyz", "Uighur", "Tatar", "Haitian Creole", "Swahili", "Hausa", "Yoruba",
+    "Zulu", "Xhosa", "Amharic", "Somali", "Tigrinya", "Shona", "Igbo", "Malagasy", "Quechua", "Aymara", "Guarani",
+    "Sundanese", "Javanese", "Filipino", "Hmong", "Fijian", "Tongan", "Samoan", "Chamorro", "Hawaiian"
+]
+languages = sorted(languages)
+
+# List of available AI models
+model_selections = [
+    'gemini-2.0-flash',
+    'gemini-2.0-pro-exp-02-05',
+    'gemini-2.0-flash-lite'
+]
+
+def translate(text, target_language, model_sel):
+    client = genai.Client(api_key=GEMINI_API_KEY)
+    prompt = f"Translate the following text into {target_language}. Only reply with the translation.\n'{text}'."
+    
+    response = client.models.generate_content(
+        model=model_sel,
+        contents=[prompt]
+    )
+    
+    return response.text
+
+# Function to summarize transcribed text
+def summarize(transcript_text, word_count, model_sel, lang_sel):
+    client = genai.Client(api_key=GEMINI_API_KEY)
+    prompt = f"Summarize the following text in {word_count} words in {lang_sel}: '{transcript_text}'."
+    
+    response = client.models.generate_content(
+        model=model_sel,
+        contents=[prompt]
+    )
+    
+    return response.text
+
+def punctuate(transcript, model_sel):
+    client = genai.Client(api_key=GEMINI_API_KEY)
+    prompt = f"Restore puncutations of the transcript from an ASR model given. Maintain the original content. Only reply with the output. text: \n{transcript}"
+
+    response = client.models.generate_content(
+        model=model_sel,
+        contents=[prompt]
+    )
+    
+    return response.text
+
+# Gradio Interface
+with gr.Blocks(theme="huggingface") as demo:
+    gr.Markdown(
+        """
+        # 📝 Persian ASR, Translation & Summarization  
+        
+        Welcome to the **Persian Speech-to-Text & NLP** platform! This app allows you to upload an audio file,  
+        get an accurate **transcription**, and enhance the output with **translation**, **summarization**,  
+        and **punctuation restoration**.
+
+        ## 🎯 How It Works  
+        1️⃣ Upload an **audio file** containing Persian speech. To transcribe YouTube videos, open this [Colab Notebook](https://colab.research.google.com/github/saeedzou/NeMo-Gradio/blob/main/yt_transcribe_gradio.ipynb)
+        2️⃣ Click **"Transcribe"** to generate the text output.  
+        3️⃣ Use additional features: **Translate**, **Summarize**, or **Restore Punctuation**.  
+        4️⃣ Customize settings: Select a **language**, **AI model**, and **summary length**.  
+        5️⃣ View and copy the processed text for your use!  
+
+        """
+    )
+
+    with gr.Row():
+        audio_input = gr.Audio(type="filepath", label="�� Upload Audio File")
+        transcript_output = gr.Textbox(label="📝 Transcription", interactive=True)
+        translation_output = gr.Textbox(label="🌍 Translation", interactive=False)
+        summarized_output = gr.Textbox(label="📖 Summarized Text", interactive=False)
+
+    transcribe_button = gr.Button("🎙️ Transcribe")
+    translate_button = gr.Button("🌐 Translate", interactive=True)
+    summarize_button = gr.Button("✂️ Summarize", interactive=True)
+    punctuate_button = gr.Button("🔤 Restore Punctuation", interactive=True)
+
+    with gr.Row():
+        word_count_input = gr.Number(value=50, label="📏 Word Count for Summary")
+        lang_selection = gr.Dropdown(choices=languages, value="English", label="🌎 Select Language")
+        model_selection = gr.Dropdown(choices=model_selections, value="gemini-2.0-flash", label="🤖 Select AI Model")
+
+    # Button Click Actions
+    transcribe_button.click(
+        transcribe, 
+        inputs=audio_input, 
+        outputs=transcript_output
+    )
+
+    translate_button.click(
+        translate, 
+        inputs=[transcript_output, lang_selection, model_selection], 
+        outputs=translation_output
+    )
+
+    summarize_button.click(
+        summarize, 
+        inputs=[transcript_output, word_count_input, model_selection, lang_selection], 
+        outputs=summarized_output
+    )
+
+    punctuate_button.click(  
+        punctuate,  
+        inputs=[transcript_output, model_selection],  
+        outputs=transcript_output  
+    )
+
+    gr.Markdown(
+        """
+        \n\n
+        ---
+
+        Powered by NVIDIA’s **NeMo Fast Conformer**, this tool is optimized for high-quality **Persian ASR (Automatic Speech Recognition)**. 
+
+        **📚 Trained on 800+ Hours of Speech Data:**  
+        - Common Voice 17 (~300 hours)  
+        - YouTube (~400 hours)  
+        - NasleMana (~90 hours)  
+        - In-house dataset (~70 hours) 
+
+        ---
+
+        ## 📜 License & Business Inquiries  
+
+        This application is licensed under **Creative Commons Attribution-NonCommercial 4.0 (CC BY-NC 4.0)**.
+        - **🛑 Non-Commercial Use Only** – Commercial use is not permitted without prior approval.  
+        - **🔗 Attribution Required** – Credit must be given to FAIM Group, Sharif University of Technology.  
+        - **❌ No Derivatives** – Modifications or adaptations of this work are not allowed.  
+
+        📜 Full License Details: [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/)  
+
+        📩 **Business Inquiries:**  
+        If you're interested in commercial applications, please contact us at:  
+        ✉️ **Email:** [[email protected]](mailto:[email protected])  
+
+        ---
+        """
+    )
+# Launch Gradio app
+demo.launch()