commit

README.md

@@ -1,13 +1,87 @@
 ---
 title: Basic STT Transcript Cleanup
-emoji: 👀
-colorFrom: red
-colorTo: red
+emoji: 🎤
+colorFrom: blue
+colorTo: green
 sdk: gradio
-sdk_version: 5.47.2
+sdk_version: 4.44.0
 app_file: app.py
 pinned: false
-short_description: Basic text transformation utility for cleaning up STT output
+short_description: Clean up speech-to-text transcripts with AI-powered text processing
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# Basic STT Transcript Cleanup Tool (Version 3)
+
+A foundational speech-to-text transcript remediation tool that provides purpose-agnostic text cleanup instructions. This is the **daily workhorse** for cleaning up raw speech-to-text transcripts that naturally contain undesirable material.
+
+## Purpose & Philosophy
+
+This tool implements **Version 3** of the Basic Speech-to-Text Cleanup prompt - a carefully crafted system prompt that provides sufficiently deterministic guidance without overstepping into actual content editing. The challenge in developing this prompt was ensuring it cleans up technical artifacts of speech-to-text conversion while preserving the authentic voice and intent of the original speaker.
+
+## Foundational Design
+
+This basic cleanup prompt serves as a **foundation layer** that can be combined with specialized text transformation prompts:
+
+- **Standalone Use**: Perfect for general transcript cleanup
+- **Modular Design**: Can be concatenated with purpose-specific prompts from extensive libraries
+- **Purpose-Agnostic**: Works across all content types and domains
+- **Extensible**: Hundreds of specialized transformation prompts can be layered on top
+
+## Features
+
+- **AI-Powered Cleanup**: Uses OpenAI's GPT models with a refined system prompt
+- **BYOK (Bring Your Own Key)**: Secure - uses your own OpenAI API key
+- **Copy to Clipboard**: Easy copying of cleaned text
+- **Re-run Capability**: Quickly re-process the same text
+- **System Prompt Viewer**: Transparent - see exactly how the AI processes your text
+- **Deterministic Processing**: Consistent, predictable cleanup results
+
+## How to Use
+
+1. **Enter API Key**: Provide your OpenAI API key (required for processing)
+2. **Paste Transcript**: Add your raw speech-to-text transcript
+3. **Process**: Click "Clean Up Transcript" to apply remediation
+4. **Copy Results**: Use the cleaned output or re-run if needed
+
+## What It Does
+
+The tool applies these **foundational improvements** to your transcripts:
+
+### Core Remediations
+- **Removes filler words** ("um", "uh", "like", etc.)
+- **Adds proper punctuation** and sentence structure
+- **Fixes obvious transcription errors** and hallucinations
+- **Removes repetitive phrases** and run-on thoughts
+- **Improves paragraph spacing** for readability
+- **Follows inferred instructions** (e.g., "scratch that from the note")
+
+### What It Preserves
+- **All important content** and meaning
+- **Original speaker's voice** and intent
+- **Factual accuracy** and details
+- **Natural flow** of conversation
+
+## Design Principles
+
+1. **Light Touch Editing**: Minimal intervention while maximizing clarity
+2. **Content Preservation**: Never removes or alters important information
+3. **Deterministic Guidance**: Consistent, predictable results
+4. **Purpose Agnostic**: Works across all content domains
+5. **Modular Foundation**: Ready for specialized prompt layering
+
+## Extended Ecosystem
+
+This basic cleanup prompt is part of a larger ecosystem:
+- **Hundreds of specialized prompts** available in shared libraries
+- **Domain-specific transformations** for various use cases
+- **Concatenation-ready design** for complex workflows
+- **Shared on Hugging Face** and other platforms
+
+## System Prompt
+
+The tool uses a carefully crafted system prompt (Version 3, September 2025) that balances cleanup effectiveness with content preservation. View the complete prompt using the "Show System Prompt" feature in the interface.
+
+## Created By
+
+**[Daniel Rosehill](https://danielrosehill.com)** - Specializing in AI-powered text processing and speech-to-text optimization workflows.
+

app.py

@@ -0,0 +1,226 @@
+import gradio as gr
+import os
+from openai import OpenAI
+# Load system prompt and README
+def load_system_prompt():
+    with open("system-prompt.md", "r") as f:
+        return f.read()
+
+SYSTEM_PROMPT_CONTENT = load_system_prompt()
+
+# README content for About tab
+README_CONTENT = """# Basic STT Transcript Cleanup Tool (Version 3)
+
+A foundational speech-to-text transcript remediation tool that provides purpose-agnostic text cleanup instructions. This is the **daily workhorse** for cleaning up raw speech-to-text transcripts that naturally contain undesirable material.
+
+## Purpose & Philosophy
+
+This tool implements **Version 3** of the Basic Speech-to-Text Cleanup prompt - a carefully crafted system prompt that provides sufficiently deterministic guidance without overstepping into actual content editing. The challenge in developing this prompt was ensuring it cleans up technical artifacts of speech-to-text conversion while preserving the authentic voice and intent of the original speaker.
+
+## Foundational Design
+
+This basic cleanup prompt serves as a **foundation layer** that can be combined with specialized text transformation prompts:
+
+- **Standalone Use**: Perfect for general transcript cleanup
+- **Modular Design**: Can be concatenated with purpose-specific prompts from extensive libraries
+- **Purpose-Agnostic**: Works across all content types and domains
+- **Extensible**: Hundreds of specialized transformation prompts can be layered on top
+
+## Features
+
+- **AI-Powered Cleanup**: Uses OpenAI's GPT models with a refined system prompt
+- **BYOK (Bring Your Own Key)**: Secure - uses your own OpenAI API key
+- **Copy to Clipboard**: Easy copying of cleaned text
+- **Re-run Capability**: Quickly re-process the same text
+- **System Prompt Viewer**: Transparent - see exactly how the AI processes your text
+- **Deterministic Processing**: Consistent, predictable cleanup results
+
+## How to Use
+
+1. **Enter API Key**: Provide your OpenAI API key (required for processing)
+2. **Paste Transcript**: Add your raw speech-to-text transcript
+3. **Process**: Click "Clean Up Transcript" to apply remediation
+4. **Copy Results**: Use the cleaned output or re-run if needed
+
+## What It Does
+
+The tool applies these **foundational improvements** to your transcripts:
+
+### Core Remediations
+- **Removes filler words** ("um", "uh", "like", etc.)
+- **Adds proper punctuation** and sentence structure
+- **Fixes obvious transcription errors** and hallucinations
+- **Removes repetitive phrases** and run-on thoughts
+- **Improves paragraph spacing** for readability
+- **Follows inferred instructions** (e.g., "scratch that from the note")
+
+### What It Preserves
+- **All important content** and meaning
+- **Original speaker's voice** and intent
+- **Factual accuracy** and details
+- **Natural flow** of conversation
+
+## Design Principles
+
+1. **Light Touch Editing**: Minimal intervention while maximizing clarity
+2. **Content Preservation**: Never removes or alters important information
+3. **Deterministic Guidance**: Consistent, predictable results
+4. **Purpose Agnostic**: Works across all content domains
+5. **Modular Foundation**: Ready for specialized prompt layering
+
+## Extended Ecosystem
+
+This basic cleanup prompt is part of a larger ecosystem:
+- **Hundreds of specialized prompts** available in shared libraries
+- **Domain-specific transformations** for various use cases
+- **Concatenation-ready design** for complex workflows
+- **Shared on Hugging Face** and other platforms
+
+## System Prompt
+
+The tool uses a carefully crafted system prompt (Version 3, September 2025) that balances cleanup effectiveness with content preservation. View the complete prompt using the "Show System Prompt" feature in the interface.
+"""
+
+def cleanup_transcript(text, api_key):
+    """Clean up STT transcript using OpenAI API"""
+    if not text.strip():
+        return "Please provide text to clean up."
+    
+    if not api_key.strip():
+        return "Please provide your OpenAI API key."
+    
+    try:
+        client = OpenAI(api_key=api_key)
+        
+        response = client.chat.completions.create(
+            model="gpt-4o-mini",  # Using cost-effective model
+            messages=[
+                {"role": "system", "content": SYSTEM_PROMPT},
+                {"role": "user", "content": text}
+            ],
+            temperature=0.3,
+            max_tokens=4000
+        )
+        
+        cleaned_text = response.choices[0].message.content
+        return cleaned_text
+        
+    except Exception as e:
+        return f"Error: {str(e)}"
+
+def copy_to_clipboard_js():
+    """JavaScript function to copy text to clipboard"""
+    return """
+    function copyToClipboard(text) {
+        navigator.clipboard.writeText(text).then(function() {
+            // You could add a toast notification here
+        });
+    }
+    """
+
+# Create the Gradio interface
+with gr.Blocks(
+    title="STT Transcript Cleanup Tool",
+    theme=gr.themes.Soft(),
+    css="""
+    .main-header {
+        text-align: center;
+        margin-bottom: 2rem;
+    }
+    .attribution {
+        text-align: center;
+        margin-top: 2rem;
+        padding: 1rem;
+        background-color: #f8f9fa;
+        border-radius: 8px;
+    }
+    """
+) as demo:
+    
+    gr.HTML("""
+    <div class="main-header">
+        <h1>🎤 STT Transcript Cleanup Tool</h1>
+        <p>Clean up speech-to-text transcripts by removing filler words, adding punctuation, and improving readability.</p>
+        <p><strong>Note:</strong> This tool requires your own OpenAI API key (BYOK - Bring Your Own Key)</p>
+    </div>
+    """)
+    
+    with gr.Tabs():
+        with gr.TabItem("Cleanup Tool"):
+            with gr.Row():
+                with gr.Column():
+                    api_key_input = gr.Textbox(
+                        label="OpenAI API Key",
+                        placeholder="sk-...",
+                        type="password",
+                        info="Your API key is not stored and only used for this session"
+                    )
+                    
+                    input_text = gr.Textbox(
+                        label="Raw STT Transcript",
+                        placeholder="Paste your speech-to-text transcript here...",
+                        lines=10,
+                        max_lines=20
+                    )
+                    
+                    with gr.Row():
+                        cleanup_btn = gr.Button("Clean Up Transcript", variant="primary")
+                        clear_btn = gr.Button("Clear", variant="secondary")
+                
+                with gr.Column():
+                    output_text = gr.Textbox(
+                        label="Cleaned Transcript",
+                        lines=10,
+                        max_lines=20,
+                        interactive=False
+                    )
+                    
+                    with gr.Row():
+                        copy_btn = gr.Button("Copy to Clipboard", variant="secondary")
+                        rerun_btn = gr.Button("Run Again", variant="secondary")
+        
+        with gr.TabItem("System Prompt"):
+            gr.Markdown("## Current System Prompt")
+            gr.Markdown("This is the prompt used to instruct the AI on how to clean up your transcripts:")
+            gr.Code(SYSTEM_PROMPT_CONTENT, language="markdown", label="system-prompt.md")
+        
+        with gr.TabItem("About"):
+            gr.Markdown(README_CONTENT)
+    
+    # Event handlers
+    cleanup_btn.click(
+        fn=cleanup_transcript,
+        inputs=[input_text, api_key_input],
+        outputs=output_text
+    )
+    
+    rerun_btn.click(
+        fn=cleanup_transcript,
+        inputs=[input_text, api_key_input],
+        outputs=output_text
+    )
+    
+    clear_btn.click(
+        fn=lambda: ("", ""),
+        outputs=[input_text, output_text]
+    )
+    
+    # Copy to clipboard functionality
+    copy_btn.click(
+        fn=None,
+        inputs=output_text,
+        outputs=None,
+        js="(text) => navigator.clipboard.writeText(text)"
+    )
+    
+    # Attribution
+    gr.HTML("""
+    <div class="attribution">
+        <p><strong>Created by:</strong> <a href="https://danielrosehill.com" target="_blank">Daniel Rosehill</a></p>
+        <p>This tool helps clean up speech-to-text transcripts by removing filler words, adding proper punctuation, 
+        and improving overall readability while preserving the original meaning and important details.</p>
+    </div>
+    """)
+
+if __name__ == "__main__":
+    demo.launch()

requirements.txt

@@ -0,0 +1,2 @@
+gradio==4.44.0
+openai==1.51.0

system-prompt.md

@@ -0,0 +1,29 @@
+# Speech To Text (STT): Basic Text Cleanup / Remediation Prompt: V3 (28/Sep/2025)
+
+You are a helpful writing assistant. 
+
+Your purpose is to lightly edit texts provided by the user. The texts which you will be receiving were generated using speech to text technology. Your function is to lightly edit the texts to improve their readability and intelligibility.
+
+Your overarching objective is to take "raw" text and reformat it for readiability. 
+
+## Editing Instructions
+
+You should apply the following set of remediations:
+
+- Remove filler words (like "um")
+- If you can infer instructions to omit certain clauses from the edited transcript you will produce (For example: "wait .. scratch that from the note") then follow the inferred instruction 
+- Add punctuation, sentence structure, and paragraph spacing'
+
+You may encounter, in the source text, obvious hallucinations or mistranscriptions. If you encounter either, infer the intended meaning rather than the version in the transcript.
+
+Beyond these basic fixes, you should edit the text lightly to omit repetitive or run on thoughts where words are repeated that would not be helpful in transcript. For example if the raw transcript read: "I want go to the cinema today to ... see ... to see ... the movie .. the movie ... Shrek" you would edit that to "I want to go to the cinema today to see the movie Shrek."
+
+Do not edit the transcript beyond these basic fixes. Assume that all details in the transcipt are important and must be preserved. 
+
+## Workflow
+
+You adhere to the following workflow:
+
+The user will provide the text (you can infer that any long text prompted carries the implicit instruction to please apply your edirts). 
+
+In response: you reply with the full remediated text without any prefixing or suffixing content or messages to the user.