import gradio as gr
import pixeltable as pxt
from pixeltable.iterators import DocumentSplitter
from pixeltable.functions import openai
import os
import requests
import tempfile
def process_document(pdf_file, api_key, voice_choice, style_choice, chunk_size, temperature, max_tokens, progress=gr.Progress()):
try:
os.environ['OPENAI_API_KEY'] = api_key
progress(0.1, desc="Initializing...")
pxt.drop_dir('document_audio', force=True)
pxt.create_dir('document_audio')
docs = pxt.create_table(
'document_audio.documents',
{
'document': pxt.DocumentType(),
'voice': pxt.StringType(),
'style': pxt.StringType()
}
)
progress(0.2, desc="Processing document...")
docs.insert([{
'document': pdf_file.name,
'voice': voice_choice,
'style': style_choice
}])
chunks = pxt.create_view(
'document_audio.chunks',
docs,
iterator=DocumentSplitter.create(
document=docs.document,
separators='token_limit',
limit=chunk_size
)
)
progress(0.4, desc="Text processing...")
chunks['content_response'] = openai.chat_completions(
messages=[
{
'role': 'system',
'content': """Transform this text segment into clear, concise content.
Structure:
1. Core concepts and points
2. Supporting details
3. Key takeaways"""
},
{'role': 'user', 'content': chunks.text}
],
model='gpt-4o-mini-2024-07-18',
max_tokens=max_tokens,
temperature=temperature
)
chunks['content'] = chunks.content_response['choices'][0]['message']['content']
progress(0.6, desc="Script generation...")
chunks['script_response'] = openai.chat_completions(
messages=[
{
'role': 'system',
'content': f"""Convert content to audio script.
Style: {docs.style}
Format:
- Clear sentence structures
- Natural pauses (...)
- Term definitions when needed
- Proper transitions
- Appropriate pronunciation guidance"""
},
{'role': 'user', 'content': chunks.content}
],
model='gpt-4o-mini-2024-07-18',
max_tokens=max_tokens,
temperature=temperature
)
chunks['script'] = chunks.script_response['choices'][0]['message']['content']
progress(0.8, desc="Audio synthesis...")
@pxt.udf(return_type=pxt.AudioType())
def generate_audio(script: str, voice: str):
if not script or not voice:
return None
try:
response = requests.post(
"https://api.openai.com/v1/audio/speech",
headers={"Authorization": f"Bearer {api_key}"},
json={"model": "tts-1", "input": script, "voice": voice}
)
if response.status_code == 200:
temp_file = tempfile.NamedTemporaryFile(delete=False, suffix='.mp3')
temp_file.write(response.content)
temp_file.close()
return temp_file.name
except Exception as e:
print(f"Error in audio synthesis: {e}")
return None
chunks['audio'] = generate_audio(chunks.script, docs.voice)
audio_path = chunks.select(chunks.audio).tail(1)['audio'][0]
results = chunks.select(chunks.content, chunks.script).collect()
display_data = [
[f"Segment {idx + 1}", row['content'], row['script']]
for idx, row in enumerate(results)
]
progress(1.0, desc="Complete")
return display_data, audio_path, "Processing complete"
except Exception as e:
return None, None, f"Error: {str(e)}"
with gr.Blocks(theme=gr.themes.Base()) as demo:
gr.Markdown(
"""
📄 Document to Audio Synthesis 🎧
"""
)
with gr.Row():
with gr.Column():
with gr.Accordion("🎯 What does it do?", open=False):
gr.Markdown("""
1. 📄 **Document Processing:** PDF extraction and token-based chunking
2. 🤖 **Content Pipeline:** LLM-powered text optimization and script generation
3. 🔊 **Audio Generation:** Neural TTS synthesis with voice modulation
""")
with gr.Column():
with gr.Accordion("⚡ How does it work?", open=False):
gr.Markdown("""
1. 📑 **Segmentation:** Token-based document chunking with configurable limits
2. 🔍 **Transformation:** Dual-pass LLM processing with temperature control
3. 🎵 **Synthesis:** OpenAI TTS with multi-voice capability
""")
gr.HTML(
"""
⚠️ API Cost Notice: This application uses OpenAI's Text-to-Speech API which incurs costs per use.
See OpenAI's TTS Documentation
for current pricing information.
"""
)
with gr.Row():
with gr.Column():
with gr.Accordion("🔑 Input & Voice", open=True):
api_key = gr.Textbox(
label="OpenAI API Key",
placeholder="sk-...",
type="password"
)
file_input = gr.File(
label="PDF Document",
file_types=[".pdf"]
)
with gr.Column():
with gr.Accordion("⚙️ Processing Configuration", open=True):
style_select = gr.Radio(
choices=["Technical", "Narrative", "Instructional", "Descriptive"],
value="Technical",
label="💫 Style"
)
with gr.Row():
voice_select = gr.Radio(
choices=["alloy", "echo", "fable", "onyx", "nova", "shimmer"],
value="onyx",
label="🎙️ Voice Model"
)
with gr.Row():
chunk_size = gr.Slider(
minimum=100, maximum=1000, value=300, step=50,
label="📏 Chunk Size"
)
temperature = gr.Slider(
minimum=0, maximum=1, value=0.7, step=0.1,
label="🌡️ Temperature"
)
max_tokens = gr.Slider(
minimum=100, maximum=1000, value=300, step=50,
label="📊 Tokens"
)
with gr.Row():
process_btn = gr.Button("🚀 Generate Audio", variant="primary", scale=2)
with gr.Tabs():
with gr.TabItem("📝 Content"):
output_table = gr.Dataframe(
headers=["🔍 Segment", "📄 Content", "🎭 Script"],
wrap=True
)
with gr.TabItem("🎧 Audio"):
with gr.Row():
with gr.Column(scale=2):
audio_output = gr.Audio(
label="🔊 Generated Audio",
type="filepath",
show_download_button=True
)
with gr.Column(scale=1):
with gr.Accordion("📚 Technical Notes", open=True):
gr.Markdown("""
- 🎯 Temperature < 0.5: Deterministic output
- 📏 Chunk size affects token context
- 🎙️ Voice models vary in prosody
- 💰 API usage is billed per character
""")
gr.HTML(
"""
"""
)
def update_interface(pdf_file, api_key, voice, style, chunk_size, temperature, max_tokens):
return process_document(
pdf_file, api_key, voice, style, chunk_size, temperature, max_tokens
)
process_btn.click(
update_interface,
inputs=[
file_input, api_key, voice_select, style_select,
chunk_size, temperature, max_tokens
],
outputs=[output_table, audio_output]
)
if __name__ == "__main__":
demo.launch()