app.py

import gradio as gr
import re
import os
import requests
import time
import soundfile as sf
import io


def show_popup():
    return gr.update(visible=True)

def hide_popup():
    return gr.update(visible=False)

def audio_to_bytes(audio):
    data, sr = sf.read(audio)
    audio_bytes = io.BytesIO()
    sf.write(audio_bytes, data, sr, format='WAV')
    audio_bytes.seek(0)
    return audio_bytes

def langswitch_API_call(audio, language):
    audio_bytes = audio_to_bytes(audio)
    files = {'file': (f'audio_chunk.wav', audio_bytes, 'audio/wav')}
    api_url = os.getenv("api_url")
    response = requests.post(f"{api_url}/online/http?language={language}", files=files)
    if response.status_code != 200:
        print(response)
        raise Exception("API error")
    return response.json()

def transcribe_base(audio, language):
    response = langswitch_API_call(audio, language)
    print(response)
    transcription = response["transcription"]
    is_new_speaker = response["is_new_speaker"]
    speaker = response["classified_speaker"]
    if is_new_speaker:
        speaker_class_string = f'New speaker detected. Assigned new ID {speaker}'
    else:
        speaker_class_string = f'Speaker found in database, ID {speaker}'
    return transcription#, speaker_class_string

def fix_italian_transcription(transcription):
    no_elision_cases = {
        "un autore", "un artista", "un uomo", "un amico", "un imperatore",
        "uno studente", "uno psicologo", "uno zio",
        "di autore", "a uomo", "su imperatore", "con amico", "per artista"
    }
    
    transcription = re.sub(r"\b(un|l|d|s|t|m|c|n|quest|all|dall|dell|nell|sull|coll|pell|dov)\s+(?=[aeiouhàèìòùáéíóú])", r"\1'", transcription)
    transcription = re.sub(r"\b(s|t|m|c|n)\s+(?=è\b|ha\b|hanno\b)", r"\1'", transcription)
    transcription = re.sub(r"\bpo\b", "po'", transcription)
    transcription = re.sub(r"\b(senz) ([aeiou])", r"\1'\2", transcription)
    transcription = transcription.replace("anch io", "anch'io")
    
    pattern_numbers = r"\b(trent|quarant|cinquant|sessant|settant|ottant|novant)\s+([aeiouàèìòù])"
    replacement_numbers = lambda m: m.group(1) + "'" + m.group(2)
    transcription = re.sub(pattern_numbers, replacement_numbers, transcription)
    
    for phrase in no_elision_cases:
        fixed = phrase.replace(" ", "'")
        transcription = transcription.replace(fixed, phrase)
    
    return transcription

def transcribe_mic(audio_microphone, language):
    print("Transcription microphone")
    transcription = transcribe_base(audio_microphone, language)
    print(transcription)
    if language=="it":
        transcription = fix_italian_transcription(transcription)
        print(transcription)
        
    return transcription
    #return transcribe_base(audio_microphone, language)

def transcribe_file(audio_upload, language):
    print("Transcription local file")
    return transcribe_base(audio_upload, language)


css_content = """
/*
.gradio-container{
    padding: 0 !important;
}
.html-container{
    padding: 0 !important;
}
*/
#orai-info{
    padding: 50px;
    text-align: center;
    font-size: 1rem;
    background: url('https://elia.eus/static/elhuyar/img/landing_page/ig.webp') rgba(0,0,0,0.8);
    background-repeat: no-repeat;
    background-position: center center;
    background-size: cover;
    background-blend-mode: multiply;
}
#orai-info-text p{
    color: white !important;
}
/*
#orai-info img{
    margin: auto;
    display: block;
    margin-bottom: 1rem;
}*/
.bold{
    font-weight: bold;
    color: inherit !important;
}
footer{
    display:none !important
}

.logos{
    display: flex;
    justify-content: center;
}
.sermas-logo{
    display: flex;
    align-items: center;
    margin-right: 3rem;
}
.sermas-logo span{
    color: white !important;
    font-size: 2.5rem;
    font-family: Verdana, Geneva, sans-serif !important;
    font-weight: bold;
}

.text-elhuyar{
    color: #0045e7;
}

#header{
    padding: 50px;
    padding-top: 30px;
    background-color: #5b65a7;
}
#header h1,h3{
    color: white;
}

button.primary{
    background-color: #5b65a7;
}
button.primary:hover{
    background-color: #3c4687;
}

button.selected{
    color: #5b65a7 !important;
}
button.selected::after{
    background-color: #5b65a7;
}

.record-button::before{
    background: #E50914;
}

"""


demo = gr.Blocks(css="""
.popup-overlay {
    position: fixed;
    top: 0; left: 0;
    width: 100vw; height: 100vh;
    background: rgba(0,0,0,0.5);
    z-index: 999;
    display: flex;
    justify-content: center;
    align-items: center;
    flex-direction: column;
}

.popup-box {
    background-color: white;
    padding: 20px;
    border-radius: 10px;
    box-shadow: 0 0 20px rgba(0,0,0,0.3);
    width: 340px;
    text-align: center;
}

.popup-button {
    background-color: #5b65a7;
    color: white;
    padding: 10px 20px;
    border-radius: 8px;
    margin-top: 10px;
    border: none;
    cursor: pointer;
}
.popup-button:hover {
    background-color: #3c4687;
}
""")


demo = gr.Blocks(css=css_content) #, fill_width=True)
with demo:
    '''
    gr.HTML("""
    <style>
        .popup-overlay {
            position: fixed;
            top: 0; left: 0;
            width: 100vw; height: 100vh;
            background: rgba(0,0,0,0.5);
            display: flex; justify-content: center; align-items: center;
            z-index: 10;
        }
        .popup-box {
            background: white;
            padding: 20px;
            border-radius: 10px;
            box-shadow: 0 0 20px rgba(0,0,0,0.2);
            text-align: center;
            width: 300px;
        }
    </style>
    """)
    '''
    
    gr.HTML("""
<div id="header">
    <h1>LANGSWITCH</h1>
    <h3>Multilingual Automatic Speech Recognition in noisy environments</h3>
</div>
""")

    with gr.Tab("Transcribe microphone"):
        iface = gr.Interface(
            fn=transcribe_mic,
            inputs=[
                gr.Audio(sources="microphone", type="filepath"),
                gr.Dropdown(label="Language", choices=[("English", "en"),
                                     ("Euskara", "eu"),
                                     ("Español", "es"),
                                     ("Français", "fr"),
                                     ("Italiano", "it")],
                            value="en")
            ],
            outputs=[
                gr.Textbox(label="Transcription", autoscroll=False),
                #gr.Textbox(label="Speaker Identification", autoscroll=False)
            ],
            allow_flagging="never",
        )

    with gr.Tab("Transcribe local file"):
        iface = gr.Interface(
            fn=transcribe_file,
            inputs=[
                gr.Audio(sources="upload", type="filepath"),
                gr.Dropdown(choices=[("English", "en"),
                                     ("Euskara", "eu"),
                                     ("Español", "es"),
                                     ("Français", "fr"),
                                     ("Italiano", "it")],
                            value="en")
            ],
            outputs=[
                gr.Textbox(label="Transcription", autoscroll=False),
                #gr.Textbox(label="Speaker Identification", autoscroll=False)
            ],
            allow_flagging="never",
        )

    ###############################
    popup = gr.Column(visible=False, elem_classes="popup-overlay")
    with popup:
        with gr.Column(elem_classes="popup-box"):
            gr.Markdown(
                """
                ### ℹ️ Notice
    
                Grabaketak ez dira gordetzen eta automatikoki ezabatzen dira.  
                No se guardan las grabaciones y se eliminan automáticamente.  
                The recordings are not saved and are automatically removed.
                """
            )
            ok_btn = gr.Button("OK", elem_classes="popup-button")
            ok_btn.click(fn=hide_popup, outputs=popup)

    # Show on load
    demo.load(fn=show_popup, outputs=popup)
    ###################
    
    gr.HTML("""
<div id="orai-info">
    <div class="logos">
        <div class="sermas-logo">
            <img src="https://sermasproject.eu/wp-content/uploads/2023/04/sermas-logo.png" width=100/>
            <span>SERMAS</span>
        </div>
        <img src="https://www.orai.eus/themes/custom/orai_for_drupal9/orai_bw.svg" width=175/>
    </div>
    <div id="orai-info-text">
        <p>The <span class="bold">LANGSWITCH</span> sub-project is part of the Open Call 1 of the <span class="bold">SERMAS</span> project. The goal of the <span class="bold">SERMAS</span> project is to provide socially-acceptable extended reality models and systems.</p>
        <p>The technology powering LANGSWITCH was developed by <span class="bold">Orai NLP Teknologiak</span></p>
        <p><span class="bold">Orai NLP Teknologiak</span> specializes in research, development, and innovation in artificial intelligence, with a focus on fostering a more competitive industrial and business landscape, enhancing public administration efficiency, and promoting a more inclusive society.</p>
    </div>
</div>
<p>""")
    
demo.queue(max_size=1)
#demo.launch(share=False, max_threads=3, auth=(os.getenv("username"), os.getenv("password")), auth_message="Please provide a username and a password.")
demo.launch(share=False, max_threads=3)