Thiwanka01's picture
Update app.py (#3)
b43ed4d verified
import gradio as gr
from transformers import pipeline
import torch
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
# Load pretrained models from Hugging Face
nlp_pipeline = pipeline("text-generation", model="gpt2") # For text generation (voice assistant)
speech_recognition_model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-large-960h")
speech_recognition_processor = Wav2Vec2Processor.from_pretrained("facebook/wav2vec2-large-960h")
# Function for voice-to-text conversion using Wav2Vec2
def recognize_speech(audio):
# Process the audio file
input_values = speech_recognition_processor(audio, return_tensors="pt").input_values
with torch.no_grad():
logits = speech_recognition_model(input_values).logits
predicted_ids = torch.argmax(logits, dim=-1)
# Decode the prediction
transcription = speech_recognition_processor.decode(predicted_ids[0])
return transcription
# Function for generating device commands using GPT-2 (e.g., for controlling smart devices)
def generate_response(user_input):
response = nlp_pipeline(user_input, max_length=50, num_return_sequences=1)[0]['generated_text']
return response
# Gradio Interface
def interact_with_system(audio=None, user_input=None):
if audio:
# Convert speech to text
transcription = recognize_speech(audio)
return transcription
elif user_input:
# Generate response to control devices
response = generate_response(user_input)
return response
else:
return "Please provide either voice or text input."
# Create a Gradio interface
interface = gr.Interface(
fn=interact_with_system,
inputs=[
gr.Audio(source="microphone", type="numpy", label="Voice Command"), # Voice input
gr.Textbox(label="Text Command") # Text input
],
outputs="text", # Output the text response (device control command)
title="AI-Driven Consumer Device Ecosystem",
description="Use voice or text commands to interact with smart devices in your ecosystem."
)
# Launch the interface
interface.launch()