Whisper-Tiny Example

This document explains how to use the QAI AppBuilder Python API to perform inference with the Whisper-Tiny speech recognition model using the Qualcomm® Hexagon™ Processor (NPU).

Supported devices

Device	SoC
Fogwise® AIRbox Q900	QCS9075

Install QAI AppBuilder

tip

1. Install QAI AppBuilder by following the QAI AppBuilder installation guide.

2. Configure ADSP environment variables as described in Create ADSP environment variables.

Run the sample

Install dependencies

Install sample dependencies in the activated virtual environment:

Device

pip3 install requests tqdm qai-hub py3-wget Pillow torch torchvision opencv-python-headless audio2numpy

Run the script

• Enter the upstream samples directory

  Device

  cd qai-appbuilder/samples

• Prepare input data (use the sample input if provided, or pass script arguments)

input audio

• Run inference

  Device

  python3 audio/Speech_Recognition/whisper_tiny_en/whisper_tiny_en.py --chipset 9075

The default input is the bundled jfk.wav. On success the terminal prints a transcription such as:

Transcription: And so my fellow Americans ask not what your country can do for you, ask what you can do for your country.

tip

The first run downloads encoder/decoder models via Qualcomm® AI Hub. Decoding needs the tiktoken vocabulary; if the device cannot reach openaipublic.blob.core.windows.net, pre-warm tiktoken.get_encoding("gpt2") on a machine with network access.

python3 run_inference.py --list
python3 run_inference.py --model whisper_tiny_en --args "--chipset 9075"