# Installation
## Requirements
Building `extract-speech` requires:
- a current stable [Rust toolchain](https://www.rust-lang.org/tools/install)
- a C/C++ build toolchain
- the Protocol Buffers compiler (`protoc`)
Install `protoc` on common platforms:
### Ubuntu and Debian
```bash
sudo apt-get update
sudo apt-get install protobuf-compiler
```
### macOS
```bash
brew install protobuf
```
### Windows
Download a `protoc` archive from the [Protocol Buffers releases](https://github.com/protocolbuffers/protobuf/releases), extract it, and add its `bin` directory to `PATH`.
## Build from source
After publication, install the optional command-line application from crates.io with:
```bash
cargo install extract-speech --features cli
```
To build the repository directly:
```bash
git clone https://github.com/RustedBytes/extract-speech.git
cd extract-speech
cargo build --release --features cli
```
The resulting executable is:
- `target/release/extract-speech` on Linux and macOS
- `target/release/extract-speech.exe` on Windows
Use `cargo build --features cli` instead when you want a faster development build. A plain `cargo build` builds the library only.
### macOS Accelerate
On Apple platforms, the optional Accelerate integration can be enabled with:
```bash
cargo build --release --features "cli,accelerate-src"
```
## Prebuilt releases
Platform CLI binaries and Python wheels can be downloaded from the project's
[GitHub releases](https://github.com/RustedBytes/extract-speech/releases). Python wheels are also
published to PyPI for CPython 3.9 through 3.15 on manylinux x86-64, macOS ARM64, and Windows
x86-64.
## Next steps
The Candle backend needs only a compatible ONNX model. The ONNX Runtime backend additionally needs a dynamic ONNX Runtime library at execution time. Continue with [Models and runtimes](models-and-runtimes.md), then see the [usage guide](usage.md).