OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Audio & Voice AI

Parakeet.cpp

Parakeet.cpp is an MIT-licensed C++ implementation for running NVIDIA Parakeet speech-recognition models on-device. It uses the Axiom tensor library, supports Metal acceleration on Apple Silicon, and exposes library and C API integration paths.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Parakeet.cpp — OpenViST project profile in Audio & Voice AI.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use it when you are building software that needs local Parakeet-based speech recognition and can integrate a C++ library or its C API. Select a model according to language, offline or streaming operation, and whether speaker diarization is needed.

Who it is for

  • C++ developers adding local speech recognition to audio-processing software
  • Developers building on-device transcription workflows for Apple Silicon
  • Teams integrating transcription from Swift, Go, Rust or Python through the C API

Documented uses

  • Transcribe audio files locally with supported Parakeet models.
  • Choose offline or streaming recognition models for English or multilingual audio.
  • Generate word timestamps and confidence scores.
  • Add voice-activity detection, batched transcription or domain-vocabulary phrase boosting.
  • Combine speech recognition with diarization to produce speaker-attributed words.
Before you choose

Requirements & limits

The repository is public and MIT-licensed. Its README describes prebuilt binaries for macOS arm64, macOS x86_64 and Linux x86_64.

  • The project is an implementation library, so using it in an application requires development and integration work.
  • Metal acceleration is available through Axiom on Apple Silicon; building Metal shaders on macOS requires the full Xcode installation.
  • Offline, streaming, language and diarization capabilities vary by selected model.
  • The documented source build requires C++20 and CMake 3.20 or later.
  • The README describes audio support for WAV, FLAC, MP3 and OGG, with automatic resampling.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Build Parakeet.cpp from source

Build the library and run its documented test command from a source checkout.

Before you start

  • C++20 with Clang 14 or later, or GCC 12 or later
  • CMake 3.20 or later
  • On macOS, macOS 13 or later and the full Xcode installation are required for Metal shader compilation

Steps

  1. Clone the repository with its recursive submodules.

  2. Change into the parakeet.cpp directory.

  3. Run the command below.

    Documented command
    make build

  4. Run the command below.

    Documented command
    make test

Expected outcome

The documented source build and test commands have been run; this guide does not claim a successful build or test result.

Guide limitations

  • Metal GPU compilation on macOS requires full Xcode, not only Command Line Tools.
  • The documented commands do not cover model conversion or transcribing an audio file.
Documentation 1 ↗
Source-based practical guide

Convert a Parakeet Nemo checkpoint to safetensors

Use the repository's documented converter to create a safetensors model file from a downloaded NVIDIA Nemo checkpoint.

Before you start

  • The Parakeet.cpp repository, including scripts/convert_nemo.py
  • huggingface-cli
  • Python packages safetensors and torch

Steps

  1. Download the *.nemo files from nvidia/parakeet-tdt_ctc-110m with the documented Hugging Face CLI command.

  2. Install safetensors and torch with pip.

  3. Run scripts/convert_nemo.py on the downloaded checkpoint and provide an output path for the safetensors file.

Expected outcome

A converted safetensors checkpoint is written to the selected output path.

Guide limitations

  • The documented example targets the 110M TDT-CTC checkpoint; other supported model types require selecting the corresponding converter model option.
  • This covers weight conversion only, not inference or application integration.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
14 October 2026
Original website title
Frikallo/parakeet.cpp · Official homepage title · retained in its source language
How we review sourcesSuggest a correction