Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -5,14 +5,51 @@ from PIL import Image
|
|
5 |
import torch
|
6 |
import os
|
7 |
import tempfile
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
8 |
|
9 |
# For TTS, try multiple options in order of preference
|
10 |
try:
|
11 |
-
# Try gTTS
|
12 |
from gtts import gTTS
|
13 |
-
|
14 |
-
|
15 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
16 |
temp_file = tempfile.NamedTemporaryFile(delete=False, suffix='.mp3')
|
17 |
temp_filename = temp_file.name
|
18 |
temp_file.close()
|
@@ -29,44 +66,34 @@ try:
|
|
29 |
os.unlink(temp_filename)
|
30 |
|
31 |
return audio_bytes, 'audio/mp3'
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
# Define alternative TTS using built-in transformers pipeline
|
37 |
-
def text2audio(story_text):
|
38 |
-
# Use a different TTS method
|
39 |
-
from transformers import pipeline
|
40 |
-
|
41 |
-
# Try a simple TTS model that should work with base transformers
|
42 |
-
synthesizer = pipeline("text-to-speech", model="facebook/mms-tts-eng")
|
43 |
-
|
44 |
-
# Generate speech
|
45 |
-
speech = synthesizer(story_text)
|
46 |
|
47 |
# Return the audio data
|
48 |
if 'audio' in speech:
|
49 |
return speech['audio'], speech.get('sampling_rate', 16000)
|
50 |
elif 'audio_array' in speech:
|
51 |
return speech['audio_array'], speech.get('sampling_rate', 16000)
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
|
56 |
-
# Simple image-to-text function
|
|
|
57 |
def img2text(image):
|
58 |
-
|
59 |
-
|
60 |
-
return
|
61 |
|
62 |
# Helper function to count words
|
63 |
def count_words(text):
|
64 |
return len(text.split())
|
65 |
|
66 |
# Improved text-to-story function without "Once upon a time" constraint
|
|
|
67 |
def text2story(text):
|
68 |
-
|
69 |
-
|
70 |
# Ask for a story without specifying how to start
|
71 |
prompt = f"""Write a children's story based on this: {text}.
|
72 |
The story should have a clear beginning, middle, and end.
|
@@ -74,7 +101,7 @@ def text2story(text):
|
|
74 |
"""
|
75 |
|
76 |
# Generate a longer text to ensure we get a complete story
|
77 |
-
story_result =
|
78 |
prompt,
|
79 |
max_length=500,
|
80 |
num_return_sequences=1,
|
@@ -160,8 +187,38 @@ def text2story(text):
|
|
160 |
|
161 |
# Basic Streamlit interface
|
162 |
st.title("Image to Audio Story")
|
163 |
-
uploaded_file = st.file_uploader("Upload an image")
|
164 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
165 |
if uploaded_file is not None:
|
166 |
# Display image
|
167 |
st.image(uploaded_file, caption="Uploaded Image")
|
@@ -169,29 +226,45 @@ if uploaded_file is not None:
|
|
169 |
# Convert to PIL Image
|
170 |
image = Image.open(uploaded_file)
|
171 |
|
172 |
-
# Image to Text
|
173 |
-
|
174 |
-
caption
|
175 |
-
|
176 |
-
|
177 |
-
|
178 |
-
|
179 |
-
|
180 |
-
|
181 |
-
|
182 |
-
|
183 |
-
st.
|
184 |
-
|
185 |
-
|
186 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
187 |
try:
|
188 |
-
audio_data, audio_format = text2audio(story)
|
189 |
-
|
190 |
-
|
191 |
-
if isinstance(audio_format, str) and audio_format.startswith('audio/'):
|
192 |
-
st.audio(audio_data, format=audio_format)
|
193 |
-
else:
|
194 |
-
st.audio(audio_data, sample_rate=audio_format)
|
195 |
except Exception as e:
|
196 |
-
|
197 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
5 |
import torch
|
6 |
import os
|
7 |
import tempfile
|
8 |
+
import time
|
9 |
+
|
10 |
+
# Use Streamlit's caching mechanisms to optimize model loading
|
11 |
+
@st.cache_resource
|
12 |
+
def load_image_to_text_pipeline():
|
13 |
+
"""Load and cache the image-to-text model"""
|
14 |
+
return pipeline("image-to-text", model="sooh-j/blip-image-captioning-base")
|
15 |
+
|
16 |
+
@st.cache_resource
|
17 |
+
def load_text_generation_pipeline():
|
18 |
+
"""Load and cache the text generation model"""
|
19 |
+
return pipeline("text-generation", model="TinyLlama/TinyLlama-1.1B-Chat-v1.0")
|
20 |
+
|
21 |
+
@st.cache_resource
|
22 |
+
def load_tts_pipeline():
|
23 |
+
"""Load and cache the text-to-speech pipeline as fallback"""
|
24 |
+
try:
|
25 |
+
return pipeline("text-to-speech", model="facebook/mms-tts-eng")
|
26 |
+
except:
|
27 |
+
# Return None if loading fails
|
28 |
+
return None
|
29 |
+
|
30 |
+
# Initialize all models at app startup
|
31 |
+
with st.spinner("Loading models (this may take a moment the first time)..."):
|
32 |
+
# Load all models at startup and cache them
|
33 |
+
img2text_model = load_image_to_text_pipeline()
|
34 |
+
story_generator_model = load_text_generation_pipeline()
|
35 |
+
tts_fallback_model = load_tts_pipeline()
|
36 |
|
37 |
# For TTS, try multiple options in order of preference
|
38 |
try:
|
39 |
+
# Try importing gTTS
|
40 |
from gtts import gTTS
|
41 |
+
has_gtts = True
|
42 |
+
except ImportError:
|
43 |
+
has_gtts = False
|
44 |
+
if tts_fallback_model is None:
|
45 |
+
st.warning("No text-to-speech capability available. Audio generation will be disabled.")
|
46 |
+
|
47 |
+
# Cache the text-to-audio conversion
|
48 |
+
@st.cache_data
|
49 |
+
def text2audio(story_text):
|
50 |
+
"""Convert text to audio with caching to avoid regenerating the same audio"""
|
51 |
+
if has_gtts:
|
52 |
+
# Use gTTS
|
53 |
temp_file = tempfile.NamedTemporaryFile(delete=False, suffix='.mp3')
|
54 |
temp_filename = temp_file.name
|
55 |
temp_file.close()
|
|
|
66 |
os.unlink(temp_filename)
|
67 |
|
68 |
return audio_bytes, 'audio/mp3'
|
69 |
+
elif tts_fallback_model is not None:
|
70 |
+
# Use transformers TTS
|
71 |
+
speech = tts_fallback_model(story_text)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
72 |
|
73 |
# Return the audio data
|
74 |
if 'audio' in speech:
|
75 |
return speech['audio'], speech.get('sampling_rate', 16000)
|
76 |
elif 'audio_array' in speech:
|
77 |
return speech['audio_array'], speech.get('sampling_rate', 16000)
|
78 |
+
|
79 |
+
# If we got here, no TTS method worked
|
80 |
+
raise Exception("No text-to-speech capability available")
|
81 |
|
82 |
+
# Simple image-to-text function using cached model
|
83 |
+
@st.cache_data
|
84 |
def img2text(image):
|
85 |
+
"""Convert image to text with caching"""
|
86 |
+
result = img2text_model(image)
|
87 |
+
return result[0]["generated_text"]
|
88 |
|
89 |
# Helper function to count words
|
90 |
def count_words(text):
|
91 |
return len(text.split())
|
92 |
|
93 |
# Improved text-to-story function without "Once upon a time" constraint
|
94 |
+
@st.cache_data
|
95 |
def text2story(text):
|
96 |
+
"""Generate a story from text with caching"""
|
|
|
97 |
# Ask for a story without specifying how to start
|
98 |
prompt = f"""Write a children's story based on this: {text}.
|
99 |
The story should have a clear beginning, middle, and end.
|
|
|
101 |
"""
|
102 |
|
103 |
# Generate a longer text to ensure we get a complete story
|
104 |
+
story_result = story_generator_model(
|
105 |
prompt,
|
106 |
max_length=500,
|
107 |
num_return_sequences=1,
|
|
|
187 |
|
188 |
# Basic Streamlit interface
|
189 |
st.title("Image to Audio Story")
|
|
|
190 |
|
191 |
+
# Add processing status indicator
|
192 |
+
status_container = st.empty()
|
193 |
+
|
194 |
+
# Initialize session state for tracking progress
|
195 |
+
if 'progress' not in st.session_state:
|
196 |
+
st.session_state.progress = {
|
197 |
+
'caption_generated': False,
|
198 |
+
'story_generated': False,
|
199 |
+
'audio_generated': False,
|
200 |
+
'caption': '',
|
201 |
+
'story': '',
|
202 |
+
'audio_data': None,
|
203 |
+
'audio_format': None
|
204 |
+
}
|
205 |
+
|
206 |
+
# File uploader
|
207 |
+
uploaded_file = st.file_uploader("Upload an image", on_change=lambda: reset_progress())
|
208 |
+
|
209 |
+
# Function to reset progress when a new file is uploaded
|
210 |
+
def reset_progress():
|
211 |
+
st.session_state.progress = {
|
212 |
+
'caption_generated': False,
|
213 |
+
'story_generated': False,
|
214 |
+
'audio_generated': False,
|
215 |
+
'caption': '',
|
216 |
+
'story': '',
|
217 |
+
'audio_data': None,
|
218 |
+
'audio_format': None
|
219 |
+
}
|
220 |
+
|
221 |
+
# Process the image if uploaded
|
222 |
if uploaded_file is not None:
|
223 |
# Display image
|
224 |
st.image(uploaded_file, caption="Uploaded Image")
|
|
|
226 |
# Convert to PIL Image
|
227 |
image = Image.open(uploaded_file)
|
228 |
|
229 |
+
# Image to Text (if not already done)
|
230 |
+
if not st.session_state.progress['caption_generated']:
|
231 |
+
status_container.info("Generating caption...")
|
232 |
+
st.session_state.progress['caption'] = img2text(image)
|
233 |
+
st.session_state.progress['caption_generated'] = True
|
234 |
+
|
235 |
+
st.write(f"Caption: {st.session_state.progress['caption']}")
|
236 |
+
|
237 |
+
# Text to Story (if not already done)
|
238 |
+
if not st.session_state.progress['story_generated']:
|
239 |
+
status_container.info("Creating story...")
|
240 |
+
st.session_state.progress['story'] = text2story(st.session_state.progress['caption'])
|
241 |
+
st.session_state.progress['story_generated'] = True
|
242 |
+
|
243 |
+
# Display word count for transparency
|
244 |
+
word_count = count_words(st.session_state.progress['story'])
|
245 |
+
st.write(f"Story ({word_count} words):")
|
246 |
+
st.write(st.session_state.progress['story'])
|
247 |
+
|
248 |
+
# Pre-generate audio in background (if not already done)
|
249 |
+
if not st.session_state.progress['audio_generated'] and (has_gtts or tts_fallback_model is not None):
|
250 |
+
status_container.info("Pre-generating audio in background...")
|
251 |
try:
|
252 |
+
st.session_state.progress['audio_data'], st.session_state.progress['audio_format'] = text2audio(st.session_state.progress['story'])
|
253 |
+
st.session_state.progress['audio_generated'] = True
|
254 |
+
status_container.success("Ready to play audio!")
|
|
|
|
|
|
|
|
|
255 |
except Exception as e:
|
256 |
+
status_container.error(f"Error pre-generating audio: {e}")
|
257 |
+
|
258 |
+
# Button to play audio
|
259 |
+
if st.button("Play the audio"):
|
260 |
+
if st.session_state.progress['audio_generated']:
|
261 |
+
# Display the audio player
|
262 |
+
if isinstance(st.session_state.progress['audio_format'], str) and st.session_state.progress['audio_format'].startswith('audio/'):
|
263 |
+
st.audio(st.session_state.progress['audio_data'], format=st.session_state.progress['audio_format'])
|
264 |
+
else:
|
265 |
+
st.audio(st.session_state.progress['audio_data'], sample_rate=st.session_state.progress['audio_format'])
|
266 |
+
else:
|
267 |
+
# Handle case where audio generation failed or is not available
|
268 |
+
st.error("Unable to play audio. Audio generation was not successful.")
|
269 |
+
else:
|
270 |
+
status_container.info("Upload an image to begin")
|