Build and run on macOS

Overview

In this section, you will build and run the audio generation application on macOS. The application uses ExecuTorch with XNNPack and Arm KleidiAI for optimized inference on Arm CPUs.

Set up the environment

Start a fresh virtual environment to avoid dependency conflicts from earlier steps:

    

        
        
cd $WORKSPACE/ML-examples/kleidiai-examples/audiogen-et/
python3.10 -m venv new-venv
source new-venv/bin/activate

    

Start a fresh virtual environment:

    

        
        
cd $WORKSPACE/ML-examples/kleidiai-examples/audiogen-et/
python3.10 -m venv new-venv
source new-venv/bin/activate

    

If you haven’t already installed ExecuTorch, install it now:

    

        
        
pip install executorch==1.0.0

    

Set model path

Set the EXECUTORCH_MODELS_PATH environment variable to the directory containing your exported ExecuTorch models:

    

        
        
export EXECUTORCH_MODELS_PATH=$WORKSPACE/ML-examples/kleidiai-examples/audiogen-et

    

This should be the directory where the .pte model files were created during the conversion step.

Build the application

Navigate to the app directory and create a build directory:

    

        
        
cd app
mkdir build && cd build

    

Run CMake to configure the build:

    

        
        
cmake ..

    

Build the application:

    

        
        
make -j8

    

The build process creates an audiogen executable in the build directory.

Download the tokenizer model

The audio generation application uses a SentencePiece-based tokenizer. Download the spiece.model file from Hugging Face:

    

        
        
curl -L https://huggingface.co/google-t5/t5-base/resolve/main/spiece.model -o $EXECUTORCH_MODELS_PATH/spiece.model

    

Verify the file was downloaded:

    

        
        
ls -lh $EXECUTORCH_MODELS_PATH/spiece.model

    

Run the application

Run the audiogen application with three required arguments:

• Model Path (-m): Directory containing your ExecuTorch models and spiece.model files
• Prompt (-p): Text description of the desired audio
• CPU Threads (-t): Number of CPU threads to use

Run the application with an example prompt:

    

        
        
./audiogen -m $EXECUTORCH_MODELS_PATH -p "warm arpeggios on house beats 120BPM with drums effect" -t 4

    

The application processes the prompt through the three model submodules and generates audio.

The output is similar to:

    

        
        I 00:00:00.002858 executorch:main.cpp:280] Resetting threadpool with num threads = 4
I 00:00:00.003228 executorch:threadpool.cpp:48] Resetting threadpool to 4 threads.
I 00:00:00.003543 executorch:main.cpp:288] Using 4 threads
I 00:00:00.003620 executorch:main.cpp:294] Model (/Users/parver01/my-workspace/ML-examples/kleidiai-examples/audiogen-et/conditioners_model.pte) loaded
I 00:00:00.003627 executorch:main.cpp:298] Model (/Users/parver01/my-workspace/ML-examples/kleidiai-examples/audiogen-et/dit_model.pte) loaded
I 00:00:00.003629 executorch:main.cpp:302] Model (/Users/parver01/my-workspace/ML-examples/kleidiai-examples/audiogen-et/autoencoder_model.pte) loaded
I 00:00:04.723761 executorch:main.cpp:478] Output saved to warm_arpeggios_on_house_beats_120bpm_with_drums_effect_99.wav
I 00:00:04.723770 executorch:main.cpp:487] T5: 72 ms
I 00:00:04.723771 executorch:main.cpp:488] DiT: 1474 ms
I 00:00:04.723772 executorch:main.cpp:489] DiT Avg per step: 184.250000 ms
I 00:00:04.723784 executorch:main.cpp:490] AutoEncoder: 3127 ms
I 00:00:04.723785 executorch:main.cpp:491] Total execution time: 4673 ms

        
    

Verify the output

If successful, the generated audio is saved as a .wav file in the current directory. The filename is based on the prompt text:

    

        
        
ls -lh warm_arpeggios_on_house_beats_120bpm_with_drums_effect_99.wav

    

You can play the audio file using any audio player on your macOS system:

    

        
        
open warm_arpeggios_on_house_beats_120bpm_with_drums_effect_99.wav

    

What you’ve accomplished and what’s next

You’ve built the audio generation application for macOS, downloaded the tokenizer model, and generated audio from a text prompt using ExecuTorch with Arm KleidiAI optimizations. The application delivers efficient inference on Arm CPUs, and lets you experiment with different prompts to generate audio.

If you want to deploy to Android instead, proceed to the next section.