Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,68 +1,72 @@
|
|
1 |
-
# import part
|
2 |
import streamlit as st
|
3 |
from transformers import pipeline
|
|
|
4 |
|
5 |
-
#
|
6 |
-
|
7 |
-
|
8 |
-
|
9 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
10 |
return text
|
11 |
|
12 |
-
# text2story -
|
13 |
def text2story(text):
|
14 |
-
#
|
15 |
-
generator =
|
16 |
|
17 |
-
#
|
18 |
-
prompt = f"
|
19 |
|
20 |
-
#
|
21 |
story_result = generator(
|
22 |
prompt,
|
23 |
-
|
24 |
num_return_sequences=1,
|
25 |
temperature=0.7,
|
26 |
-
top_k=
|
27 |
-
top_p=0.
|
28 |
do_sample=True
|
29 |
)
|
30 |
|
31 |
-
# Extract
|
32 |
story_text = story_result[0]['generated_text']
|
33 |
story_text = story_text.replace(prompt, "Once upon a time, ")
|
34 |
|
35 |
-
# Find a natural ending point
|
36 |
-
|
37 |
-
if
|
38 |
-
|
39 |
-
shortened_text = " ".join(words[:100])
|
40 |
-
|
41 |
-
# Find the last complete sentence
|
42 |
-
last_period = shortened_text.rfind('.')
|
43 |
-
last_question = shortened_text.rfind('?')
|
44 |
-
last_exclamation = shortened_text.rfind('!')
|
45 |
-
|
46 |
-
# Find the last sentence ending punctuation
|
47 |
-
last_end = max(last_period, last_question, last_exclamation)
|
48 |
-
|
49 |
-
if last_end > 0:
|
50 |
-
# Truncate at the end of the last complete sentence
|
51 |
-
story_text = shortened_text[:last_end + 1]
|
52 |
-
else:
|
53 |
-
# If no sentence ending found, just use the shortened text
|
54 |
-
story_text = shortened_text
|
55 |
|
56 |
return story_text
|
57 |
|
58 |
-
# text2audio -
|
59 |
def text2audio(story_text):
|
60 |
try:
|
61 |
-
#
|
62 |
-
synthesizer =
|
63 |
|
64 |
-
#
|
65 |
-
max_chars =
|
66 |
if len(story_text) > max_chars:
|
67 |
last_period = story_text[:max_chars].rfind('.')
|
68 |
if last_period > 0:
|
@@ -72,46 +76,57 @@ def text2audio(story_text):
|
|
72 |
|
73 |
# Generate speech
|
74 |
speech = synthesizer(story_text)
|
75 |
-
|
76 |
-
# Get output information
|
77 |
-
st.write(f"Speech output keys: {list(speech.keys())}")
|
78 |
-
|
79 |
return speech
|
80 |
|
81 |
except Exception as e:
|
82 |
st.error(f"Error generating audio: {str(e)}")
|
83 |
return None
|
84 |
|
85 |
-
# main
|
86 |
-
st.set_page_config(page_title="
|
87 |
-
st.header("
|
88 |
-
uploaded_file = st.file_uploader("Select an Image...")
|
89 |
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
|
|
101 |
|
102 |
-
#
|
103 |
-
st.
|
104 |
-
story = text2story(caption)
|
105 |
-
st.write(story)
|
106 |
|
107 |
-
#
|
108 |
-
|
109 |
-
speech_output = text2audio(story)
|
110 |
|
111 |
-
#
|
112 |
-
if st.button("
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
113 |
if speech_output is not None:
|
114 |
-
# Try to play the audio directly
|
115 |
try:
|
116 |
if 'audio' in speech_output and 'sampling_rate' in speech_output:
|
117 |
st.audio(speech_output['audio'], sample_rate=speech_output['sampling_rate'])
|
@@ -120,21 +135,15 @@ if uploaded_file is not None:
|
|
120 |
elif 'waveform' in speech_output and 'sample_rate' in speech_output:
|
121 |
st.audio(speech_output['waveform'], sample_rate=speech_output['sample_rate'])
|
122 |
else:
|
123 |
-
# Try
|
124 |
for key, value in speech_output.items():
|
125 |
if hasattr(value, '__len__') and len(value) > 1000:
|
126 |
-
|
127 |
-
|
128 |
-
elif 'sample_rate' in speech_output:
|
129 |
-
st.audio(value, sample_rate=speech_output['sample_rate'])
|
130 |
-
elif 'sampling_rate' in speech_output:
|
131 |
-
st.audio(value, sample_rate=speech_output['sampling_rate'])
|
132 |
-
else:
|
133 |
-
st.audio(value, sample_rate=24000) # Default sample rate
|
134 |
break
|
135 |
else:
|
136 |
-
st.error(
|
137 |
except Exception as e:
|
138 |
st.error(f"Error playing audio: {str(e)}")
|
139 |
else:
|
140 |
-
st.error("Audio generation failed
|
|
|
1 |
+
# import part
|
2 |
import streamlit as st
|
3 |
from transformers import pipeline
|
4 |
+
from PIL import Image
|
5 |
|
6 |
+
# Set global caching options for Transformers
|
7 |
+
from transformers import set_caching_enabled
|
8 |
+
set_caching_enabled(True)
|
9 |
+
|
10 |
+
# function part with caching for better performance
|
11 |
+
@st.cache_resource
|
12 |
+
def load_image_captioning_model():
|
13 |
+
return pipeline("image-to-text", model="sooh-j/blip-image-captioning-base")
|
14 |
+
|
15 |
+
@st.cache_resource
|
16 |
+
def load_text_generator():
|
17 |
+
return pipeline("text-generation", model="TinyLlama/TinyLlama-1.1B-Chat-v1.0")
|
18 |
+
|
19 |
+
@st.cache_resource
|
20 |
+
def load_tts_model():
|
21 |
+
return pipeline("text-to-speech", model="HelpingAI/HelpingAI-TTS-v1")
|
22 |
+
|
23 |
+
# img2text - Using the original model with more constraints
|
24 |
+
def img2text(image):
|
25 |
+
# Load the model (cached)
|
26 |
+
image_to_text = load_image_captioning_model()
|
27 |
+
|
28 |
+
# Strongly limit output length for speed
|
29 |
+
text = image_to_text(image, max_new_tokens=15)[0]["generated_text"]
|
30 |
return text
|
31 |
|
32 |
+
# text2story - Much more constrained for speed
|
33 |
def text2story(text):
|
34 |
+
# Load the model (cached)
|
35 |
+
generator = load_text_generator()
|
36 |
|
37 |
+
# Very brief prompt to minimize work
|
38 |
+
prompt = f"Short story about {text}: Once upon a time, "
|
39 |
|
40 |
+
# Very constrained parameters for maximum speed
|
41 |
story_result = generator(
|
42 |
prompt,
|
43 |
+
max_new_tokens=60, # Much shorter output
|
44 |
num_return_sequences=1,
|
45 |
temperature=0.7,
|
46 |
+
top_k=10, # Lower value = faster
|
47 |
+
top_p=0.9, # Lower value = faster
|
48 |
do_sample=True
|
49 |
)
|
50 |
|
51 |
+
# Extract and clean text
|
52 |
story_text = story_result[0]['generated_text']
|
53 |
story_text = story_text.replace(prompt, "Once upon a time, ")
|
54 |
|
55 |
+
# Find a natural ending point
|
56 |
+
last_period = story_text.rfind('.')
|
57 |
+
if last_period > 30: # Ensure we have at least some content
|
58 |
+
story_text = story_text[:last_period + 1]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
59 |
|
60 |
return story_text
|
61 |
|
62 |
+
# text2audio - Minimal text for faster processing
|
63 |
def text2audio(story_text):
|
64 |
try:
|
65 |
+
# Load the model (cached)
|
66 |
+
synthesizer = load_tts_model()
|
67 |
|
68 |
+
# Aggressively limit text length to speed up TTS
|
69 |
+
max_chars = 200 # Much shorter than before
|
70 |
if len(story_text) > max_chars:
|
71 |
last_period = story_text[:max_chars].rfind('.')
|
72 |
if last_period > 0:
|
|
|
76 |
|
77 |
# Generate speech
|
78 |
speech = synthesizer(story_text)
|
|
|
|
|
|
|
|
|
79 |
return speech
|
80 |
|
81 |
except Exception as e:
|
82 |
st.error(f"Error generating audio: {str(e)}")
|
83 |
return None
|
84 |
|
85 |
+
# Streamlined main UI
|
86 |
+
st.set_page_config(page_title="Image to Story", page_icon="📚")
|
87 |
+
st.header("Image to Audio Story")
|
|
|
88 |
|
89 |
+
# Add info about processing time
|
90 |
+
st.info("Note: Processing may take some time as the models are loading. Please be patient.")
|
91 |
+
|
92 |
+
# Cache the file uploader state
|
93 |
+
if "uploaded_file" not in st.session_state:
|
94 |
+
st.session_state["uploaded_file"] = None
|
95 |
|
96 |
+
uploaded_file = st.file_uploader("Select an Image...", key="file_uploader")
|
97 |
+
|
98 |
+
# Process the image if uploaded
|
99 |
+
if uploaded_file is not None:
|
100 |
+
st.session_state["uploaded_file"] = uploaded_file
|
101 |
|
102 |
+
# Display the uploaded image
|
103 |
+
st.image(uploaded_file, caption="Uploaded Image", use_column_width=True)
|
|
|
|
|
104 |
|
105 |
+
# Convert to PIL image
|
106 |
+
image = Image.open(uploaded_file)
|
|
|
107 |
|
108 |
+
# Optional processing toggle to let user decide
|
109 |
+
if st.button("Generate Story and Audio"):
|
110 |
+
col1, col2 = st.columns(2)
|
111 |
+
|
112 |
+
# Stage 1: Image to Text with minimal output
|
113 |
+
with col1:
|
114 |
+
with st.spinner('Captioning image...'):
|
115 |
+
caption = img2text(image)
|
116 |
+
st.write(f"**Caption:** {caption}")
|
117 |
+
|
118 |
+
# Stage 2: Text to Story with minimal length
|
119 |
+
with col2:
|
120 |
+
with st.spinner('Creating story...'):
|
121 |
+
story = text2story(caption)
|
122 |
+
st.write(f"**Story:** {story}")
|
123 |
+
|
124 |
+
# Stage 3: Audio with minimal text
|
125 |
+
with st.spinner('Generating audio...'):
|
126 |
+
speech_output = text2audio(story)
|
127 |
+
|
128 |
+
# Display audio immediately
|
129 |
if speech_output is not None:
|
|
|
130 |
try:
|
131 |
if 'audio' in speech_output and 'sampling_rate' in speech_output:
|
132 |
st.audio(speech_output['audio'], sample_rate=speech_output['sampling_rate'])
|
|
|
135 |
elif 'waveform' in speech_output and 'sample_rate' in speech_output:
|
136 |
st.audio(speech_output['waveform'], sample_rate=speech_output['sample_rate'])
|
137 |
else:
|
138 |
+
# Try any array-like data
|
139 |
for key, value in speech_output.items():
|
140 |
if hasattr(value, '__len__') and len(value) > 1000:
|
141 |
+
sample_rate = speech_output.get('sampling_rate', speech_output.get('sample_rate', 24000))
|
142 |
+
st.audio(value, sample_rate=sample_rate)
|
|
|
|
|
|
|
|
|
|
|
|
|
143 |
break
|
144 |
else:
|
145 |
+
st.error("Could not find audio data in the output")
|
146 |
except Exception as e:
|
147 |
st.error(f"Error playing audio: {str(e)}")
|
148 |
else:
|
149 |
+
st.error("Audio generation failed")
|