Skip to content

README.md

Latest commit

 

History

History
55 lines (40 loc) · 1.78 KB

README.md

File metadata and controls

55 lines (40 loc) · 1.78 KB

Speech Recognition — Hailo-8/8L/10H

Whisper speech-to-text that runs on all Hailo accelerators (Hailo-8, Hailo-8L, Hailo-10H).

Record from your microphone and get real-time transcription, or transcribe audio files.

Unlike simple_whisper_chat (H10-only, uses hailo_platform.genai.Speech2Text), this app uses the low-level HailoRT InferModel API with separate encoder/decoder HEFs — compatible with all Hailo devices.

Prerequisites

  • Hailo-8, Hailo-8L, or Hailo-10H accelerator
  • Python 3.10+, HailoRT 4.20+
  • ffmpeg and libportaudio2: sudo apt install ffmpeg libportaudio2

Installation

pip install -e ".[speech-rec]"

Models (HEF files and decoder assets) are managed by the repo's central resource system (resources_config.yaml) and auto-downloaded on first run via resolve_hef_paths(). No manual download step is needed.

Usage

Live microphone recording (interactive loop):

python -m hailo_apps.python.standalone_apps.speech_recognition.speech_recognition

Press Enter to start recording, Enter again to stop. Press 'q' to quit.

Transcribe an audio file:

python -m hailo_apps.python.standalone_apps.speech_recognition.speech_recognition \
    --audio /path/to/audio.wav

Options

Flag Description
--audio PATH Transcribe a file instead of recording
--arch {hailo8,hailo8l,hailo10h} Target architecture (auto-detected if omitted)
--variant {base,tiny,tiny.en} Whisper variant (default: base)
--duration N Max recording length in seconds (default: 10)
--list-models List available models and exit

Supported Models

Variant Hailo-8 Hailo-8L Hailo-10H
base ✓ ✓ ✓
tiny ✓ ✓ ✓
tiny.en — — ✓