Assignment1 / app.py
CR7CAD's picture
Update app.py
7df9b81 verified
raw
history blame
4.96 kB
# import part
import streamlit as st
from transformers import pipeline
import os
import numpy as np
import io
import scipy.io.wavfile as wavfile
# function part
# img2text
def img2text(image_path):
image_to_text = pipeline("image-to-text", model="sooh-j/blip-image-captioning-base")
text = image_to_text(image_path)[0]["generated_text"]
return text
# text2story
def text2story(text):
# Using a smaller text generation model
generator = pipeline("text-generation", model="TinyLlama/TinyLlama-1.1B-Chat-v1.0")
# Create a prompt for the story generation
prompt = f"Write a fun children's story based on this: {text}. Once upon a time, "
# Generate the story
story_result = generator(
prompt,
max_length=150,
num_return_sequences=1,
temperature=0.7,
top_k=50,
top_p=0.95,
do_sample=True
)
# Extract the generated text
story_text = story_result[0]['generated_text']
story_text = story_text.replace(prompt, "Once upon a time, ")
# Make sure the story is at least 100 words
words = story_text.split()
if len(words) > 100:
# Simply truncate to 100 words
story_text = " ".join(words[:100])
return story_text
# text2audio - REVISED to use facebook/mms-tts-eng model
def text2audio(story_text):
try:
# Use a smaller and more reliable TTS model
synthesizer = pipeline("text-to-speech", model="facebook/mms-tts-eng")
# Break the text into smaller chunks if needed (prevent timeout)
max_chunk_size = 200 # characters
chunks = []
for i in range(0, len(story_text), max_chunk_size):
chunk = story_text[i:i+max_chunk_size]
# Make sure we break at word boundaries
if i+max_chunk_size < len(story_text) and story_text[i+max_chunk_size] != ' ':
# Find the last space in this chunk
last_space = chunk.rfind(' ')
if last_space != -1:
chunk = chunk[:last_space]
chunks.append(chunk)
# Process each chunk
audio_arrays = []
sampling_rate = None
for chunk in chunks:
if not chunk.strip(): # Skip empty chunks
continue
speech = synthesizer(chunk)
if sampling_rate is None:
sampling_rate = speech["sampling_rate"]
audio_arrays.append(speech["audio"])
# Combine all audio chunks
combined_audio = np.concatenate(audio_arrays)
# Create a BytesIO object to store the wave file
wav_buffer = io.BytesIO()
wavfile.write(wav_buffer, sampling_rate, combined_audio)
wav_buffer.seek(0) # Rewind the buffer
return {
"audio": wav_buffer.getvalue(),
"sampling_rate": sampling_rate
}
except Exception as e:
st.error(f"Error generating audio: {str(e)}")
# Fallback to a pre-recorded audio file if available
try:
with open("fallback_audio.wav", "rb") as f:
return {
"audio": f.read(),
"sampling_rate": 22050 # Common sample rate
}
except:
return None
# Function to save temporary image file
def save_uploaded_image(uploaded_file):
if not os.path.exists("temp"):
os.makedirs("temp")
image_path = os.path.join("temp", uploaded_file.name)
with open(image_path, "wb") as f:
f.write(uploaded_file.getvalue())
return image_path
# main part
st.set_page_config(page_title="Your Image to Audio Story", page_icon="🦜")
st.header("Turn Your Image to Audio Story")
uploaded_file = st.file_uploader("Select an Image...")
if uploaded_file is not None:
# Display the uploaded image
st.image(uploaded_file, caption="Uploaded Image", use_container_width=True)
# Save the image temporarily
image_path = save_uploaded_image(uploaded_file)
# Stage 1: Image to Text
st.text('Processing img2text...')
caption = img2text(image_path)
st.write(caption)
# Stage 2: Text to Story
st.text('Generating a story...')
story = text2story(caption)
st.write(story)
# Stage 3: Story to Audio data
st.text('Generating audio data...')
audio_data = text2audio(story)
# Play button
if st.button("Play Audio"):
if audio_data:
st.audio(
audio_data["audio"],
format="audio/wav",
start_time=0,
sample_rate=audio_data["sampling_rate"]
)
else:
st.error("Failed to generate audio. Please try again.")
# Clean up the temporary file
try:
os.remove(image_path)
except:
pass