#whisper-mbt

#Prerequisites

• MoonBit toolchain
• CMake, C/C++ compiler
• (macOS) Xcode Command Line Tools

#Quick Start (standalone)

git clone --recursive https://github.com/mizchi/whisper-mbt.git
cd whisper-mbt

# Build whisper.cpp and download model
just setup

# Run
WHISPER_MODEL=models/ggml-base.bin \
WHISPER_WAV=vendor/whisper.cpp/samples/jfk.wav \
  moon run src/main --target native

#Use as a dependency

#1. Add the package

moon add mizchi/whisper-mbt

{
  "name": "your/project",
  "deps": {
    "mizchi/whisper-mbt": {
      "path": "/path/to/whisper-mbt"
    }
  },
  "preferred-target": "native"
}

#2. Build whisper.cpp

git clone https://github.com/ggml-org/whisper.cpp.git
cd whisper.cpp

# macOS (Metal)
cmake -B build -DCMAKE_BUILD_TYPE=Release \
  -DGGML_METAL=ON -DGGML_METAL_EMBED_LIBRARY=ON \
  -DBUILD_SHARED_LIBS=OFF \
  -DWHISPER_BUILD_EXAMPLES=OFF -DWHISPER_BUILD_TESTS=OFF -DWHISPER_BUILD_SERVER=OFF
cmake --build build --config Release -j$(sysctl -n hw.logicalcpu)

# Linux (CPU only)
cmake -B build -DCMAKE_BUILD_TYPE=Release \
  -DBUILD_SHARED_LIBS=OFF \
  -DWHISPER_BUILD_EXAMPLES=OFF -DWHISPER_BUILD_TESTS=OFF -DWHISPER_BUILD_SERVER=OFF
cmake --build build --config Release -j$(nproc)

#3. Configure your main package

// cmd/main/moon.pkg
import {
  "mizchi/whisper-mbt" @whisper,
}

options(
  "is-main": true,
  link: {
    "native": {
      // macOS (Metal + Accelerate)
      "cc-link-flags": "/path/to/whisper.cpp/build/src/libwhisper.a /path/to/whisper.cpp/build/ggml/src/libggml.a /path/to/whisper.cpp/build/ggml/src/libggml-base.a /path/to/whisper.cpp/build/ggml/src/libggml-cpu.a /path/to/whisper.cpp/build/ggml/src/ggml-metal/libggml-metal.a /path/to/whisper.cpp/build/ggml/src/ggml-blas/libggml-blas.a -lstdc++ -framework Accelerate -framework Metal -framework Foundation -framework MetalKit",
    },
  },
  "supported-targets": [ "native" ],
)

"cc-link-flags": "/path/to/whisper.cpp/build/src/libwhisper.a /path/to/whisper.cpp/build/ggml/src/libggml.a /path/to/whisper.cpp/build/ggml/src/libggml-base.a /path/to/whisper.cpp/build/ggml/src/libggml-cpu.a -lstdc++ -lm -lpthread"

#4. Download a model

# ggml-base (147 MB)
curl -L -o models/ggml-base.bin \
  https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.bin

#5. Use

fn main {
  let ctx = @whisper.WhisperContext::init("models/ggml-base.bin")
  match ctx {
    None => println("Failed to load model")
    Some(ctx) => {
      let segments = ctx.transcribe("audio.wav", language="auto")
      for i = 0; i < segments.length(); i = i + 1 {
        println(segments[i].text)
      }
      ctx.free()
    }
  }
}

#API

#WhisperContext

WhisperContext::init(model_path : String) -> WhisperContext?
WhisperContext::transcribe(self, wav_path, language?="en", translate?=false, n_threads?=4, ...) -> Array[Segment]
WhisperContext::transcribe_parallel(self, wav_path, n_processors?=4, ...) -> Array[Segment]
WhisperContext::get_tokens(self, segment_index) -> Array[TokenData]
WhisperContext::model_info(self) -> ModelInfo
WhisperContext::detected_language(self) -> String
WhisperContext::get_timings(self) -> Timings
WhisperContext::print_timings(self) -> Unit
WhisperContext::free(self) -> Unit

#transcribe options

#VAD (Voice Activity Detection)

curl -L -o models/ggml-silero-v5.1.2.bin \
  https://huggingface.co/ggml-org/whisper-vad/resolve/main/ggml-silero-v5.1.2.bin

let segments = ctx.transcribe(
  "audio.wav",
  language="auto",
  vad_model_path="models/ggml-silero-v5.1.2.bin",
)

#Parallel inference

let segments = ctx.transcribe_parallel(
  "long_audio.wav",
  n_processors=4,
  language="auto",
)

#Updating vendored headers

just build-whisper
just update-headers

#License