audioqna / app.py
avfranco's picture
Update app.py
42907f8 verified
import os
import spaces
import torch
import gradio as gr
from transformers import pipeline
MODEL_NAME = "openai/whisper-large-v3"
BATCH_SIZE = 8
FILE_LIMIT_MB = 1000
device = 0 if torch.cuda.is_available() else "cpu"
pipe = pipeline(
task="automatic-speech-recognition",
model=MODEL_NAME,
chunk_length_s=30,
device=device,
)
def respond_to_question_llama(transcript, question):
from huggingface_hub import InferenceClient
client = InferenceClient(
"meta-llama/Meta-Llama-3.1-70B-Instruct",
token=os.environ["HUGGINGFACEHUB_API_TOKEN"],
)
response = client.chat_completion(
messages=[{"role": "user", "content": f"Transcript: {transcript}\n\nUser: {question}"}],
max_tokens=4096,
).choices[0].message.content
return response
@spaces.GPU
def audio_transcribe(inputs):
status=True
text="Arquivo de audio nao carregado!"
status=False
if inputs is None:
raise gr.Error("No audio file submitted! Please upload an audio file before submitting your request.")
else:
text = pipe(inputs, batch_size=BATCH_SIZE, return_timestamps=True)['text']
status = True
return [text, gr.Textbox(visible=status),gr.Textbox(visible=status),gr.Textbox(visible=status)]
def hidden_ask_question():
return [gr.Textbox(visible=False),gr.Textbox(visible=False),gr.Textbox(visible=False)]
with gr.Blocks() as transcriberUI:
gr.Markdown(
"""
# Ola!
Clique no botao abaixo para selecionar o Audio que deseja conversar!
Ambiente disponivel 24x7. Running on ZeroGPU with openai/whisper-large-v3
"""
)
inp = gr.Audio(sources="upload", type="filepath", label="Audio file")
transcribe = gr.Textbox(label="Transcricao", show_label=True, show_copy_button=True)
ask_question = gr.Textbox(label="Ask a question", visible=False)
response_output = gr.Textbox(label="Response", visible=False)
submit_question = gr.Button("Submit question", visible=False)
submit_button = gr.Button("Transcribe to Chat", variant='primary', size='sm')
clear_button = gr.ClearButton([transcribe,response_output,inp, ask_question])
def ask_question_callback(transcription,question):
if ask_question:
response = respond_to_question_llama(transcription, question)
else:
response = "No question asked"
return response
#inp.upload(audio_transcribe, inputs=inp, outputs=[transcribe,ask_question,submit_question, response_output])
submit_button.click(audio_transcribe, inputs=inp, outputs=[transcribe,ask_question,submit_question, response_output])
submit_question.click(ask_question_callback, outputs=[response_output], inputs=[transcribe, ask_question])
clear_button.click(hidden_ask_question,outputs=[ask_question,response_output,submit_question])
transcriberUI.queue().launch()