app.py

import gradio as gr
import librosa
from asr import transcribe, ASR_EXAMPLES, ASR_NOTE
from lid import identify  # Import Language Identification model

# Function to detect language and transcribe speech
def auto_transcribe(audio):
    # Detect language
    detected_lang = identify(audio)
    
    # Debug: Print detected language
    print(f"Detected Language: {detected_lang}")  

    # Ensure the detected language is Swahili or English
    if detected_lang.startswith("swh"):
        detected_lang = "swh"
    elif detected_lang.startswith("eng"):
        detected_lang = "eng"
    else:
        return f"Error: Detected language '{detected_lang}' is not supported."

    # Transcribe using detected language
    return transcribe(audio, lang=detected_lang)

# Speech-to-Text Interface with Auto Language Detection
mms_transcribe = gr.Interface(
    fn=auto_transcribe,
    inputs=gr.Audio(),
    outputs="text",
    examples=ASR_EXAMPLES,
    title="Speech-to-Text (Auto Language Detection)",
    description="Automatically detects whether speech is in Swahili or English and transcribes it.",
    article=ASR_NOTE,
    allow_flagging="never",
)

# Main Gradio App
with gr.Blocks() as demo:
    gr.Markdown("<p align='center' style='font-size: 20px;'>MMS Speech-to-Text</p>")
    gr.HTML("<center>Automatically detects and transcribes Swahili or English speech.</center>")

    mms_transcribe.render()

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