Gradio Text-to-Speech to Subtitle Demo

This demo application demonstrates how to use OpenAI's Text-to-Speech API combined with Speechhall's Speech-to-Text API to generate subtitles for audio content.

Features

• Text Input: Enter any text you want to convert to speech
• Voice Selection: Choose from 6 different OpenAI TTS voices
• Audio Generation: Convert text to high-quality speech audio
• Audio Playback: Play the generated audio directly in the browser
• Subtitle Generation: Transcribe the audio to SRT subtitle format
• Model Selection: Choose from different speech-to-text models
• Language Support: Support for multiple languages

Screenshot

Setup

1. Install Dependencies

# Using uv (recommended)
uv sync

# Or using pip
pip install -e .

2. Set Environment Variables

You need API keys for both services:

export OPENAI_API_KEY="your-openai-api-key"
export SPEECHALL_API_TOKEN="your-speechall-api-token"

Or create a .env file in the project root:

OPENAI_API_KEY=your-openai-api-key
SPEECHALL_API_TOKEN=your-speechall-api-token

3. Run the Demo

python gradio_demo.py

The application will start on http://localhost:7860

How to Use

1. Enter Text: Type or paste the text you want to convert to speech
2. Choose Voice: Select from available OpenAI TTS voices (alloy, echo, fable, onyx, nova, shimmer)
3. Generate Speech: Click "🎵 Generate Speech" to create the audio
4. Play Audio: Use the audio player to listen to the generated speech
5. Choose Model: Select a speech-to-text model for transcription
6. Generate Subtitles: Click "📝 Generate Subtitles" to create SRT format subtitles
7. View Results: The SRT subtitle content will appear in the text area

Use Cases

This demo is perfect for:

• Content Creators: Generate subtitles for voice-over content
• Video Production: Create synchronized subtitles for TTS narration
• Accessibility: Generate captions for synthetic speech
• Quality Testing: Compare original text with transcribed results
• Workflow Testing: Test the complete TTS → STT pipeline

Main Script Examples

The main.py script provides comprehensive examples of using the Speechall API directly without the Gradio interface. It demonstrates various transcription scenarios and API features.

Features Demonstrated

• API Client Setup: Authentication and configuration
• Model Discovery: List all available speech-to-text models
• Local File Transcription: Transcribe audio files from your local system
• Remote URL Transcription: Transcribe audio from web URLs
• Advanced Features: Speaker diarization, custom vocabulary, smart formatting
• Multiple Output Formats: JSON and plain text responses
• Error Handling: Comprehensive error handling and user feedback

Running the Examples

1. Set your API token:

   export SPEECHALL_API_TOKEN="your-speechall-api-token"

2. Update the audio file path in the script (line 241):

   local_audio_file = os.path.expanduser("~/Downloads/your-audio-file.mp3")

3. Run the script:

   python main.py

Example Scenarios

The script includes five different transcription examples:

1. List Available Models: Discover all speech-to-text models and their capabilities
2. Basic Local Transcription: Transcribe a local audio file with custom language and model settings
3. Remote URL Transcription: Transcribe audio directly from a web URL with word-level timestamps
4. Advanced Features: Use speaker diarization, custom vocabulary, and smart formatting
5. Default Options: Simple transcription with minimal configuration

This script is perfect for:

• Learning the API: Understanding different features and options
• Testing Integration: Verify API connectivity and functionality
• Batch Processing: Process multiple audio files programmatically
• Feature Exploration: Try different models and advanced features

API Information

• OpenAI TTS: Uses the tts-1 model for fast, high-quality speech generation
• Speechhall STT: Supports multiple models including AssemblyAI, OpenAI Whisper, and Deepgram
• Output Format: Generates proper SRT subtitle format with timestamps

Troubleshooting

• Missing API Keys: Make sure both environment variables are set correctly
• Network Issues: Check your internet connection and API quotas
• Audio Issues: Ensure your browser supports MP3 audio playback
• Large Text: Very long text may take time to process or hit API limits