File size: 10,165 Bytes
461f36f
ddd522e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
df10cdd
ddd522e
df10cdd
 
651c6b8
ddd522e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f477d87
 
ddd522e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
461f36f
 
1c0e277
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
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(
        """
        <div>
            <img src="https://raw.githubusercontent.com/pixeltable/pixeltable/main/docs/source/data/pixeltable-logo-large.png" alt="Pixeltable" style="max-width: 200px; margin-bottom: 20px;" />
            <h1 style="margin-bottom: 0.5em;">📄 Document to Audio Synthesis 🎧</h1>
        </div>
        """
    )

    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(
        """
        <div style="background-color: #FFF3CD; border: 1px solid #FFEEBA; padding: 1rem; margin: 1rem 0; border-radius: 4px;">
            <p style="margin: 0; color: #856404;">
                ⚠️ <strong>API Cost Notice:</strong> This application uses OpenAI's Text-to-Speech API which incurs costs per use. 
                See <a href="https://platform.openai.com/docs/guides/text-to-speech" target="_blank" style="color: #856404; text-decoration: underline;">OpenAI's TTS Documentation</a> 
                for current pricing information.
            </p>
        </div>
        """
    )

    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(
        """
        <div style="text-align: center; margin-top: 1rem; padding-top: 1rem; border-top: 1px solid #ccc;">
            <p style="margin: 0; color: #666; font-size: 0.8em;">
                🚀 Powered by <a href="https://github.com/pixeltable/pixeltable" target="_blank" style="color: #F25022; text-decoration: none;">Pixeltable</a>
                | 📚 <a href="https://docs.pixeltable.com" target="_blank" style="color: #666;">Docs</a>
                | 🤗 <a href="https://huggingface.co/spaces/Pixeltable" target="_blank" style="color: #666;">HF Space</a>
            </p>
        </div>
        """
    )
    
    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()