Skip to content

README.md

Latest commit

 

History

History
128 lines (91 loc) · 3.8 KB

README.md

File metadata and controls

128 lines (91 loc) · 3.8 KB

Hailo Simple Whisper Chat Example

A simple example application demonstrating audio transcription using Hailo's Whisper speech-to-text model.

Features

  • Audio file loading and processing
  • Speech-to-text transcription
  • Segment-based output
  • Default audio file: audio.wav (in same directory)
  • Auto-downloads model on first run

Prerequisites

  • Hailo AI accelerator device (H10 or compatible)
  • Python 3.10+
  • Hailo Platform SDK

Installation

Linux

Before running this example, ensure GenAI dependencies are installed:

# From the repository root directory
pip install -e ".[gen-ai]"

Windows

Use a clean virtual environment before installing the dependencies:

py -m venv wind_venv
.\wind_venv\Scripts\Activate.ps1
pip install .\hailort-<version>-cp<python>-cp<python>-win_amd64.whl

# From the repository root directory
pip install -e ".[gen-ai]"

This will install required packages including:

  • NumPy

For complete installation instructions, see: GenAI Applications Installation Guide

Files

  • simple_whisper_chat.py - Main example script demonstrating Whisper transcription functionality
  • audio.wav - Sample audio file for testing

Usage

Run the example with the default audio file:

python -m hailo_apps.python.gen_ai_apps.simple_whisper_chat.simple_whisper_chat

Or specify a custom audio file:

python -m hailo_apps.python.gen_ai_apps.simple_whisper_chat.simple_whisper_chat --audio /path/to/your/audio.wav

Command Line Options

  • --hef-path PATH - Specify a custom path to the HEF model file
  • --list-models - List available models for this application
  • --audio PATH - Path to audio file (default: audio.wav in same directory)

Example Output

The example will:

  1. Initialize the Hailo device and load the Whisper model
  2. Load the audio file (default: audio.wav in the same directory)
  3. Process the audio and generate transcription segments
  4. Display the complete transcription
  5. Clean up resources

Model Requirements

The example uses the WHISPER_MODEL_NAME_H10 model which is automatically downloaded on first run. No manual download required.

Note: Models are downloaded automatically when you run the example for the first time.

Audio File Format

The example supports WAV files with:

  • Sample rate: Any (will be processed by the model)
  • Channels: Mono or stereo (will be processed)
  • Sample width: 16-bit PCM

Troubleshooting

Model not found error

  • Ensure Hailo models are properly installed
  • Check model paths in the resource directory
  • The model will be downloaded automatically on first run

Device initialization failed

  • Verify Hailo device is connected and recognized
  • Check device permissions

Audio file not found

  • Verify the audio file exists at the specified path
  • If using default, ensure audio.wav is in the same directory as the script
  • Check file permissions

Error reading WAV file

  • Ensure the file is a valid WAV format
  • Check that the file is not corrupted
  • Verify the file is readable

Import errors

  • Ensure Hailo Platform SDK is properly installed
  • Verify Python environment has all required packages (NumPy)

How it works

The example demonstrates basic Whisper transcription:

  1. Creates a VDevice for Hailo hardware access
  2. Initializes a Speech2Text instance with the Whisper model
  3. Loads the audio file using Python's wave module
  4. Converts audio to the format expected by the model (float32, normalized)
  5. Generates transcription segments using the Whisper model
  6. Combines segments into a complete transcription
  7. Cleans up resources properly

This is a simplified example. For more advanced features like real-time audio recording, streaming transcription, and voice interaction, see the full Voice Assistant application.