app.py

import speech_recognition as sr
import gradio as gr

# 🎯 Speech to Text function with chunking for large audio
def speech_to_text(audio_file):
    r = sr.Recognizer()
    text_result = []
    with sr.AudioFile(audio_file) as source:
        while True:
            # প্রতি 60 সেকেন্ডে ভাগ করা হচ্ছে
            audio = r.record(source, duration=60)  
            if not audio.get_raw_data():  # ডেটা না থাকলে লুপ শেষ
                break
            try:
                chunk_text = r.recognize_google(audio, language="en-US")
                text_result.append(chunk_text)
            except sr.UnknownValueError:
                text_result.append("❌ [Unrecognized audio]")
            except sr.RequestError as e:
                text_result.append(f"⚠️ API Error: {e}")
                break
    return " ".join(text_result).strip()

# Gradio wrapper
def stt_gradio(audio):
    if audio is None:
        return "No audio provided"
    return speech_to_text(audio)

# Gradio UI
with gr.Blocks() as demo:
    gr.Markdown("## 🎤 English Speech to Text (Short & Long Audio Supported)")
    audio_input = gr.Audio(
        label="🎙️ Record or Upload Audio", 
        type="filepath", 
        sources=["microphone", "upload"],  # ✅ মাইক্রোফোন + আপলোড
        format="wav",
        streaming=False
    )
    text_output = gr.Textbox(label="📝 Transcribed Text", lines=10)
    submit_btn = gr.Button("Convert")
    submit_btn.click(fn=stt_gradio, inputs=audio_input, outputs=text_output)

if __name__ == "__main__":
    demo.launch()