---
title: SpeakEdge - AI Sales Communication Coach
emoji: 🎯
colorFrom: purple
colorTo: blue
sdk: gradio
sdk_version: 5.49.1
app_file: app.py
pinned: false
license: mit
hardware: zero-gpu
---

# 🎯 SpeakEdge - Professional Communication Practice Platform

**Master your sales conversations with AI-powered roleplay and comprehensive feedback**

SpeakEdge is an advanced platform designed to help sales professionals practice and perfect their communication skills through realistic AI-powered roleplay scenarios. Get detailed feedback on your structure, delivery, and listening skills.

## ✨ Features

### 🎭 Realistic Roleplay Scenarios
Practice with 9 different sales situations:
- Cold call to a big shot CEO
- Cold call to the secretary of a big shot CEO
- Inbound customer discovery call
- Demo run through with a customer
- Negotiating pricing with a customer who wants unreasonable discounts
- Renewal or churn reduction with an angry customer
- Upselling to an existing customer
- Sharing update about a good week with your Head of Sales
- Sharing update about a poor week with your Head of Sales

### 🤖 AI-Powered Conversation
- **Speech-to-Text (STT)**: Whisper large-v3 for accurate transcription
- **Text-to-Speech (TTS)**: Parler-TTS with accent customization
- **LLM**: Mistral-7B-Instruct for dynamic, context-aware responses

### 📊 Comprehensive Feedback Analysis

**Structure and Clarity**
- Logical Flow: Ideas follow coherent sequence
- Simplicity of Language: Avoids jargon and complexity
- Conciseness: Eliminates unnecessary verbosity

**Delivery and Presence**
- Pace & Pauses: Optimal speaking speed (WPM)
- Confidence & Projection: Strong, definitive language
- Energy Level: Appropriate intensity for context

**Listening and Interaction**
- Active Listening: Questions and curiosity
- Acknowledgment: Validates others' points
- Reflection: Paraphrases and mirrors

### 📈 Advanced Metrics
- Words Per Minute (WPM)
- Filler words count and percentage
- Weak words detection
- Repetition analysis
- Unique AI insights (e.g., "23% of sentences started with 'And'")

### 🎨 Beautiful Dashboard
Professional, visual feedback dashboard with:
- Score breakdowns
- Color-coded performance indicators
- Actionable recommendations
- Unique pattern recognition

## 🚀 Quick Start

### Prerequisites
- Python 3.9+
- CUDA-capable GPU (recommended) or CPU
- Hugging Face account with token

### Installation

1. **Clone the repository**
```bash
git clone https://github.com/yourusername/SpeakEdge.git
cd SpeakEdge
```

2. **Install dependencies**
```bash
pip install -r requirements.txt
```

3. **Set up Hugging Face token**
```bash
export HF_TOKEN=your_hugging_face_token_here
```

Or create a `.env` file:
```
HF_TOKEN=your_hugging_face_token_here
```

4. **Run the application**
```bash
python app.py
```

The app will launch on `http://localhost:7860`

## 🌐 Hugging Face Spaces Deployment

This application is optimized for Hugging Face Spaces with **Zero GPU** (Dynamic H200 GPU Allocation).

### Deploy to HF Spaces

1. **Create a new Space**
   - Go to [Hugging Face Spaces](https://huggingface.co/spaces)
   - Click "Create new Space"
   - Choose Gradio as the SDK
   - Enable GPU (Zero GPU)

2. **Upload files**
   - Upload all Python files (`app.py`, `models.py`, `scenarios.py`, `feedback_analyzer.py`, `dashboard.py`, `config.py`)
   - Upload `requirements.txt`
   - Copy content from `README_HF_SPACE.md` to the Space's README.md

3. **Set secrets (optional)**
   - Go to Space settings
   - Add `HF_TOKEN` as a secret (optional but recommended for better rate limits)

4. **Configure Space**
   - The app will automatically start
   - Zero GPU will dynamically allocate H200 when needed

## 🎮 How to Use

1. **Select Scenario**: Choose the sales situation you want to practice
2. **Customize Bot**: 
   - Select accent (American, British, Australian, Indian, Neutral)
   - Choose personality (Professional, Friendly, Skeptical, etc.)
   - Give your bot a name
3. **Start Roleplay**: Click "Start Roleplay" to begin
4. **Engage**: Speak naturally using your microphone
5. **Interact**: The AI bot will respond with voice and text
6. **End Session**: Click "End Roleplay" when finished
7. **Review Feedback**: Analyze your comprehensive performance dashboard

## 📁 Project Structure

```
SpeakEdge/
├── app.py                  # Main Gradio application
├── models.py              # STT, TTS, LLM model management
├── scenarios.py           # Scenario definitions and prompts
├── feedback_analyzer.py   # Feedback analysis logic
├── dashboard.py           # HTML dashboard generation
├── config.py              # Configuration settings
├── requirements.txt       # Python dependencies
├── README.md             # This file
└── README_HF_SPACE.md    # README for Hugging Face Spaces
```

## 🛠️ Technical Details

### Models Used

**Speech-to-Text**
- Model: `openai/whisper-large-v3`
- Accuracy: High
- Latency: ~2-3 seconds per audio clip

**Text-to-Speech**
- Model: `parler-tts/parler-tts-mini-v1`
- Features: Multi-accent support
- Quality: Natural-sounding voices

**Language Model**
- Model: `mistralai/Mistral-7B-Instruct-v0.3`
- Purpose: Dynamic conversation & feedback generation
- Context: Last 6 messages for coherence
- Advantage: No approval needed, excellent performance

### Performance Optimization

- **Zero GPU Integration**: Uses `@spaces.GPU` decorator for efficient GPU allocation
- **Lazy Loading**: Models load only when needed
- **Mixed Precision**: FP16 for faster inference on GPU
- **Model Caching**: Reuses loaded models across requests

## 🎯 Feedback Categories

### Structure and Clarity (3 sub-aspects)
Analyzes logical flow, language simplicity, and conciseness of communication.

### Delivery and Presence (3 sub-aspects)
Evaluates pace, confidence, and energy in speech delivery.

### Listening and Interaction (3 sub-aspects)
Assesses active listening, acknowledgment, and adaptive responses.

## 💡 Unique Features

- **Pattern Recognition**: Identifies sentence starter patterns
- **Repetition Detection**: Highlights overused words
- **Buzzword Tracking**: Warns about jargon overuse
- **Contextual Insights**: AI-generated observations specific to your conversation
- **Scenario-Specific Feedback**: Tailored to the sales situation

## 🔧 Configuration

Edit `config.py` to customize:
- Model selections
- Generation parameters
- Audio settings
- Feedback thresholds

## 🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

## 📄 License

This project is licensed under the MIT License.

## 🙏 Acknowledgments

- OpenAI Whisper for STT
- Parler-TTS for multi-accent TTS
- Mistral AI for Mistral-7B-Instruct model
- Hugging Face for hosting and Zero GPU infrastructure

## 📞 Support

For issues, questions, or suggestions, please open an issue on GitHub.

---

**Made with ❤️ for sales professionals looking to elevate their communication game**