Skip to content

quickstart.md

Latest commit

 

History

History
154 lines (100 loc) · 3.92 KB

quickstart.md

File metadata and controls

154 lines (100 loc) · 3.92 KB

Quick Start

This guide helps you go from a clean checkout to a working speech transcription using the STT-Runner library.

The steps below assume a Linux or macOS development environment.

Table of Contents

What you will build

Output Description
whisper-cli Command-line tool that runs speech-to-text transcription on a WAV file.
stt-cpp-tests Unit test binary for validating the library (optional).

Prerequisites

Minimum requirements:

  • CMake 3.27+
  • Python 3.9+
  • C++ compiler toolchain
    • GCC / Clang (Linux)
    • Xcode toolchain (macOS)

Optional platform tools:

Supported Platforms & CMake presets

The supported build platforms and CMake presets matrix is given below. The CMake presets (aka build targets) are given in the first column and the build platform in the first row. For example, native builds are tested on Linux-x86_64, Linux-aarch64, and macOS-aarch64. The x-android-aarch64 preset (targets Android devices running on aarch64) is tested on Linux-x86_64 and macOS-aarch64.

CMake preset / Host Platform Linux-x86_64 Linux-aarch64 macOS-aarch64 Android
native ✅ ✅ ✅ -
x-android-aarch64 ✅ - ✅ -
x-linux-aarch64 ✅ ✅ † - -

† Use native preset.

Steps

1) Clone the repository

git clone https://github.com/Arm-Examples/STT-Runner.git
cd STT-Runner/

2) Download the model and test assets

This repository uses a small default Whisper model for validation.

Downloads are performed automatically during CMake configuration, but you can prefetch them once per checkout:

python scripts/py/download_resources.py

Downloaded assets will appear in:

resources_downloaded/

Model notes:

  • The default model is ggml-base.en and is not quantized.
  • To reduce compute cost, you can use quantized models such as Q4_0 via the whisper.cpp quantization tool.
  • The default model configuration is declared in scripts/py/requirements.json.

3) Configure a native build

Configure the project using the native CMake preset and enable the example executable:

cmake -B build --preset=native -DBUILD_EXECUTABLE=ON

To also build the test suite:

cmake -B build --preset=native -DBUILD_EXECUTABLE=ON -DBUILD_UNIT_TESTS=ON

4) Build the project

cmake -j --build build

This produces the binaries under:

build/bin/

5) Run a transcription

Run the example CLI with a model and audio file:

./build/bin/whisper-cli \
    -m resources_downloaded/models/model.bin \
    /path/to/audio/audiofile.wav

The tool will output the transcription directly to the console.

If you do not have a WAV file available, you can generate one using any tool that exports 16-bit PCM WAV.

Next Steps