app.py

import gradio as gr
import os
import requests
from transformers import pipeline

# Set your FastAPI backend endpoint
BACKEND_URL = "https://35d2-41-84-202-90.ngrok-free.app/submit-feedback"

# Map of models
model_map = {
    "english": "jonatasgrosman/wav2vec2-large-xlsr-53-english"
}

# Create storage directory
os.makedirs("responses", exist_ok=True)

# Transcription function
def transcribe(audio, language):
    asr = pipeline("automatic-speech-recognition", model=model_map[language], device=0)
    text = asr(audio)["text"]
    return text, audio

# Save feedback by sending it to FastAPI backend
def save_feedback(audio_file, transcription, age_group, gender, evaluated_language, speak_level, write_level,
                  native, native_language, env, device, domain, accuracy, orthography, meaning, errors,
                  performance, improvement, usability, technical_issues, final_comments, email):
    try:
        # Read binary content of audio file
        with open(audio_file, "rb") as f:
            audio_content = f.read()

        # Prepare metadata as form fields
        metadata = {
            "transcription": transcription,
            "age_group": age_group,
            "gender": gender,
            "evaluated_language": evaluated_language,
            "speak_level": speak_level,
            "write_level": write_level,
            "native": native,
            "native_language": native_language,
            "environment": env,
            "device": device,
            "domain": domain,
            "accuracy": accuracy,
            "orthography": orthography,
            "meaning": meaning,
            "errors": ",".join(errors) if errors else "",
            "performance": performance,
            "improvement": improvement,
            "usability": usability,
            "technical_issues": technical_issues,
            "final_comments": final_comments,
            "email": email
        }

        files = {
            "audio_file": ("audio.wav", audio_content, "audio/wav")
        }

        response = requests.post(BACKEND_URL, data=metadata, files=files, timeout=20)

        if response.status_code == 201:
            return "✅ Feedback submitted successfully. Thank you!"
        else:
            return f"⚠️ Submission failed: {response.status_code} — {response.text}"

    except Exception as e:
        return f"❌ Could not connect to the backend: {str(e)}"


# Gradio UI
with gr.Blocks() as demo:
    gr.Markdown("## African ASR + Feedback")

    with gr.Row():
        audio_input = gr.Audio(sources=["upload", "microphone"], type="filepath", label="Upload or record audio")
        lang = gr.Dropdown(list(model_map.keys()), label="Select Language")

    transcribed_text = gr.Textbox(label="Transcribed Text")
    submit_btn = gr.Button("Transcribe")
    submit_btn.click(fn=transcribe, inputs=[audio_input, lang], outputs=[transcribed_text, audio_input])

    gr.Markdown("---\n## Feedback Form")

    age_group = gr.Dropdown(["18 to 30", "31 to 50", "50+", "Prefer not to say"], label="Age Group")
    gender = gr.Dropdown(["Male", "Female", "Prefer not to say", "Other"], label="Gender")
    evaluated_language = gr.Dropdown(list(model_map.keys()), label="Which language did you evaluate for?")
    speak_level = gr.Slider(1, 10, label="How well do you speak this language?")
    write_level = gr.Slider(1, 10, label="How well do you write the language?")
    native = gr.Radio(["Yes", "No"], label="Are you a native speaker of this language?")
    native_language = gr.Textbox(label="If not, what is your native language?")
    env = gr.Dropdown(["Studio/Professional Recording", "Quiet Room", "Noisy Background", "Multiple Environments", "Unsure", "Other"], label="Recording environment")
    device = gr.Dropdown(["Mobile Phone/Tablet", "Tablet", "Laptop/Computer Microphone", "Dedicated Microphone", "Unsure", "Other"], label="Recording device")
    domain = gr.Textbox(label="Was the speech related to a specific domain or topic? (Optional)")
    accuracy = gr.Slider(1, 10, label="How accurate was the model’s transcription?")
    orthography = gr.Dropdown(["Yes, mostly correct", "No, major issues", "Partially", "Not Applicable"], label="Did the transcription use standard orthography?")
    meaning = gr.Slider(1, 10, label="Did the transcription preserve the original meaning?")
    errors = gr.CheckboxGroup([
        "Substitutions", "Omissions", "Insertions", "Pronunciation-related", "Diacritic Errors",
        "Code-switching Errors", "Named Entity Errors", "Punctuation Errors", "No significant errors"
    ], label="Which errors were prominent?")
    performance = gr.Textbox(label="What did the model do well? What did it struggle with?")
    improvement = gr.Textbox(label="How could this ASR model be improved?")
    usability = gr.Slider(1, 5, label="How easy was it to use the tool?")
    technical_issues = gr.Textbox(label="Did you encounter any technical issues?")
    final_comments = gr.Textbox(label="Any other comments or suggestions?")
    email = gr.Textbox(label="Email (optional)")

    save_btn = gr.Button("Submit Feedback")
    output_msg = gr.Textbox(interactive=False)
    save_btn.click(fn=save_feedback, 
                   inputs=[audio_input, transcribed_text, age_group, gender, evaluated_language, speak_level, write_level,
                           native, native_language, env, device, domain, accuracy, orthography, meaning, errors,
                           performance, improvement, usability, technical_issues, final_comments, email],
                   outputs=[output_msg])

# Launch the interface
demo.launch()