app.py

import json
import gradio as gr
import requests
import os

def is_file_larger_than_30mb(file_path):
    try:
        file_size = os.path.getsize(file_path)
        return file_size > (300* 1024 * 1024)
    except FileNotFoundError:
        return False
    except PermissionError:
        return False
    except Exception as e:
        return False

def upload_audio(audio_path):
    try:
        size = is_file_larger_than_30mb(audio_path)
        if size == True:
            return 'size'
        with open(audio_path, 'rb') as audio_file:
            response = requests.post('http://sls-titan-5.csail.mit.edu:8080/upload/', files={'audio_file': audio_file})
        if response.status_code == 200:
            return response.json()["path"]
    except:
        return None

def upload_audio_13b(audio_path):
    try:
        size = is_file_larger_than_30mb(audio_path)
        if size == True:
            return 'size'
        with open(audio_path, 'rb') as audio_file:
            response = requests.post('http://sls-titan-7.csail.mit.edu:8080/upload/', files={'audio_file': audio_file})
        if response.status_code == 200:
            return response.json()["path"]
    except:
        return None

def predict(audio_path_m, audio_path_t, question, model):
    if ((audio_path_m is None) != (audio_path_t is None)) == False:
        return "Please upload and only upload one recording, either upload the audio file or record using microphone.", "Please upload and only upload one recording, either upload the audio file or record using microphone."
    else:
        audio_path = audio_path_m or audio_path_t
    if model == '7B (Default)':
        upload_statues = upload_audio(audio_path)
        if upload_statues == None:
            return 'Please upload an audio file.'
        if upload_statues == 'size':
            return 'This demo does not support audio file size larger than 30MB.'
        if question == '':
            return 'Please ask a question.'
        print(audio_path, question)
        response = requests.put('http://sls-titan-5.csail.mit.edu:8080/items/0', json={
            'audio_path': audio_path, 'question': question
        })
        answer_7b = json.loads(response.content)
        ans_str_7b = answer_7b['output']
        return ans_str_7b

    if model == '13B (Beta)':
        upload_statues = upload_audio_13b(audio_path)
        if upload_statues == None:
            return 'Please upload an audio file.'
        if upload_statues == 'size':
            return 'This demo does not support audio file size larger than 30MB.'
        if question == '':
            return 'Please ask a question.'
        print(audio_path, question)
        response = requests.put('http://sls-titan-7.csail.mit.edu:8080/items/0', json={
            'audio_path': audio_path, 'question': question
        })
        answer_13b = json.loads(response.content)
        ans_str_13b = answer_13b['output']
        return ans_str_13b

if __name__ == '__main__':
    link = "https://github.com/YuanGongND/ltu"
    text = "[Github]"
    paper_link = "https://arxiv.org/pdf/2309.14405.pdf"
    paper_text = "[ASRU Paper]"
    sample_audio_link = "https://drive.google.com/drive/folders/17yeBevX0LIS1ugt0DZDOoJolwxvncMja?usp=sharing"
    sample_audio_text = "[sample audios from AudioSet evaluation set]"
    demo = gr.Interface(fn=predict,
                        inputs=[gr.Audio(type="filepath", source='microphone', label='Please either upload an audio file or record using the microphone.', show_label=True), gr.Audio(type="filepath"),
                                gr.Textbox(value='What can be inferred from the spoken text and sounds? Why?', label='Edit the textbox to ask your own questions!'),
                                gr.Radio(["7B (Default)", "13B (Beta)"], value='7B (Default)', label="LLM size", info="All experiments in the ASRU paper are 7B LLM.")],
                        outputs=[gr.Textbox(label="LTU-AS-Output")],
                        cache_examples=True,
                        title="Demo of LTU-AS",
                        description="LTU-AS an improved version of LTU. LTU-AS is stronger in spoken text understanding and music understanding. " + f"<a href='{paper_link}'>{paper_text}</a> <br>" +
                                    "LTU-AS is authored by Yuan Gong, Alexander H. Liu, Hongyin Luo, Leonid Karlinsky, and James Glass (MIT & MIT-IBM Watson AI Lab). <br>" +
                                    "Input should be wav file sampled at 16kHz. This demo trims input audio to 10 seconds. <br>" +
                                    "Code of LTU-AS will be available soon at " + f"<a href='{link}'>{text}</a> <br>" +
                                    "**Research Demo, Not for Commercial Use (Due to license of LLaMA).**")
    demo.launch(debug=False, share=False)