Market map
Open models and engines
Speech recognition models and engines anyone can run. We track 64 of them: each one researched on its maker's own pages, with every figure read from an official source and linked to where it came from.
8of 64
run on Mac
3of 64
run on Windows
4of 64
run on Linux
5of 64
on iPhone or iPad
4of 64
on Android
4of 64
say speech stays on the device
63of 64
are open source
3of 64
are discontinued
Counted from the 64 entries below, as each maker's own pages list them.
All 64
- Whisper
OpenAI · Self-hosted
An open-source speech recognition model from OpenAI that transcribes and translates speech in many languages and can be run locally.
- NVIDIA Parakeet
NVIDIA · Self-hosted
NVIDIA's Parakeet family of speech recognition models, released with open weights and run through the NeMo toolkit.
- Voxtral
Mistral AI · Self-hosted
Mistral AI's Voxtral speech models, released with open weights for transcription and audio understanding.
- NVIDIA Canary
NVIDIA · Self-hosted
NVIDIA's Canary family of open speech recognition models, some of which also translate speech.
- faster-whisper
SYSTRAN · Self-hosted
A reimplementation of OpenAI's Whisper on the CTranslate2 engine that transcribes faster and with less memory.
- Distil-Whisper
Hugging Face · Self-hosted
Smaller, faster distilled versions of OpenAI's Whisper from Hugging Face for English speech recognition.
- Moonshine
Moonshine AI · Mac, Windows, Linux, iPhone, Android
Open speech-to-text models and a voice library from Moonshine AI for running transcription on the device.
- wav2vec 2.0
Meta AI · Self-hosted
Meta's self-supervised speech representation model, whose fine-tuned checkpoints are widely used for speech recognition.
- IBM Granite Speech
IBM · Self-hosted
IBM's Granite Speech models, released with open weights for speech recognition and speech translation.
- Meta MMS
Meta AI · Self-hosted
Meta's Massively Multilingual Speech project, which provides open speech recognition models covering a very large number of languages.
- FunASR
ModelScope · Self-hosted
An open-source speech recognition toolkit from ModelScope for training and running models such as Paraformer and Fun-ASR.
- Omnilingual ASR
Meta AI · Self-hosted
Meta's open-source multilingual speech recognition models designed to cover a very large number of languages.
- Vosk
Alpha Cephei · Android, Linux, Self-hosted
An offline open-source speech recognition toolkit with models for many languages, for apps on small devices and servers.
- MLX Whisper
Apple MLX team · Mac
Whisper speech recognition for Apple silicon through the MLX framework, published in Apple's MLX examples.
- SpeechBrain
SpeechBrain · Self-hosted
An open-source PyTorch toolkit for speech tasks, including speech recognition.
- OWSM
Carnegie Mellon University WAVLab · Self-hosted
Open Whisper-style Speech Models from Carnegie Mellon University's WAVLab, trained on public data and released through ESPnet.
- SeamlessM4T
Meta AI · Self-hosted
Meta's multilingual and multimodal translation models that also handle speech recognition.
- sherpa-onnx
Next-gen Kaldi team · Mac, Windows, Linux, iPhone, Android
An open-source toolkit from the next-generation Kaldi project that runs speech recognition and other speech tasks locally with ONNX Runtime.
- Cohere Transcribe
Cohere · Self-hosted
Cohere's open speech recognition model, released with downloadable weights for enterprises and developers building transcription workloads.
- GigaAM
SaluteDevices (Sber) · Self-hosted
Open speech recognition models from the SaluteDevices team at Sber, including Russian and multilingual versions.
- Kotoba-Whisper
Kotoba Technologies · Self-hosted
Distilled Whisper models for Japanese speech recognition from Kotoba Technologies.
An open-source speech-to-text engine from Mozilla based on Baidu's Deep Speech research and built on TensorFlow, now discontinued.
- Parakeet MLX
Senstella · Mac
An implementation of NVIDIA's Parakeet models for Apple silicon using the MLX framework.
- Qwen3-ASR
Alibaba Cloud (Qwen team) · Self-hosted
An open-source series of speech recognition models from the Qwen team at Alibaba Cloud.
- Breeze ASR
MediaTek Research · Self-hosted
MediaTek Research's speech recognition models fine-tuned from Whisper, with support for Mandarin and English mixed speech.
- CMU Sphinx (PocketSphinx)
Carnegie Mellon University · Self-hosted
A lightweight speech recognizer from the CMU Sphinx project, a family of open-source speech recognition systems.
An open-source speech-to-text toolkit continued from Mozilla's DeepSpeech by Coqui, which says it is no longer actively maintained.
- CrisperWhisper
Nyra Health · Self-hosted
A Whisper-based model that transcribes speech verbatim, including filler words, with word-level timestamps.
- ESPnet
ESPnet developers · Self-hosted
An open-source end-to-end speech processing toolkit for research, covering speech recognition and related tasks.
- FluidAudio
Fluid Inference · Mac, iPhone, iPad
A Swift SDK from Fluid Inference that runs speech models such as Parakeet on Apple devices through Core ML.
- ivrit.ai Whisper
ivrit.ai · Self-hosted
Whisper models fine-tuned for Hebrew speech recognition by the ivrit.ai project.
- Kaldi
Kaldi contributors · Self-hosted
An open-source speech recognition toolkit written in C++ and aimed at researchers.
- Kyutai STT
Kyutai · Self-hosted
Streaming speech-to-text models released with open weights by the open-science AI lab Kyutai.
- NVIDIA Nemotron Speech ASR
NVIDIA · Self-hosted
NVIDIA's Nemotron streaming speech recognition models, released with open weights on Hugging Face.
- onnx-asr
istupakov · Self-hosted
A lightweight Python package for running speech recognition models with ONNX Runtime.
- PaddleSpeech
PaddlePaddle · Self-hosted
An open-source speech toolkit on PaddlePaddle that includes streaming speech recognition.
- Qwen2-Audio
Alibaba Cloud (Qwen team) · Self-hosted
Alibaba's Qwen audio-language model, released with open weights, that takes speech and audio as input and can transcribe it.
- SenseVoice
FunAudioLLM team · Self-hosted
An open speech model for recognizing Mandarin, Cantonese, English, Japanese and Korean speech.
- SpeechRecognition (Python)
Anthony Zhang (Uberi) · Self-hosted
A Python library that wraps several speech recognition engines and web APIs behind one interface.
- Typhoon ASR
SCB 10X (Typhoon) · Self-hosted
Speech recognition models for Thai from the Typhoon team at SCB 10X, including a fine-tune of Whisper large.
- whisper.cpp
ggml-org · Mac, Windows, Linux, iPhone, Android
A plain C and C++ port of OpenAI's Whisper that runs speech recognition on a wide range of devices without heavy dependencies.
- WhisperKit
Argmax · Mac, iPhone, iPad
An open-source Swift toolkit from Argmax that runs Whisper and other speech models on Apple devices.
- WhisperLive
Collabora · Self-hosted
A nearly live implementation of OpenAI's Whisper from Collabora for streaming transcription.
- WhisperLiveKit
Quentin Fuxa · Self-hosted
An open-source tool for real-time, local speech-to-text with streaming recognition.
- WhisperX
Max Bain · Self-hosted
An open-source toolkit built on Whisper that adds word-level timestamps and speaker diarization to transcripts.
- GLM-ASR
Z.ai · Self-hosted
An open-source speech recognition model family from Z.ai, including the compact GLM-ASR-Nano.
- IndicConformer
AI4Bharat · Self-hosted
AI4Bharat's open multilingual speech recognition models for Indian languages.
- insanely-fast-whisper
Vaibhav Srivastav · Self-hosted
A command-line tool for fast Whisper transcription on local hardware, built on Hugging Face Transformers.
- k2 and icefall
Next-gen Kaldi team · Self-hosted
The training side of the next-generation Kaldi project: k2 provides differentiable finite-state algorithms and icefall provides speech recognition recipes built on them.
- KB-Whisper
KBLab, National Library of Sweden · Self-hosted
Whisper models fine-tuned for Swedish speech recognition by KBLab at the National Library of Sweden.
- Lightning Whisper MLX
Mustafa Aljadery · Mac
A Whisper implementation for Apple silicon built on the MLX framework.
- NB-Whisper
National Library of Norway AI Lab · Self-hosted
Whisper models fine-tuned for Norwegian speech recognition by the National Library of Norway's AI lab.
- NVIDIA NeMo
NVIDIA · Self-hosted
NVIDIA's open-source toolkit for building, training and running conversational AI models, including speech recognition.
- NVIDIA Riva
NVIDIA · Self-hosted
NVIDIA's speech AI SDK for deploying speech recognition and speech synthesis services on your own infrastructure.
- OLMoASR
Allen Institute for AI · Self-hosted
English speech recognition models, with open training data and code, from the Allen Institute for AI.
- Phi-4-multimodal
Microsoft · Self-hosted
Microsoft's compact multimodal model in the Phi family, with open weights, that accepts speech and can transcribe it.
- PhoWhisper
VinAI Research · Self-hosted
Whisper models fine-tuned for Vietnamese speech recognition by VinAI Research.
- ReazonSpeech
Reazon Research · Self-hosted
An open Japanese speech corpus and speech recognition models from Reazon Research.
- Reverb ASR
Rev · Self-hosted
Rev's open speech recognition model, released with inference code and weights.
- VibeVoice-ASR
Microsoft Research · Self-hosted
A speech-to-text model from Microsoft Research for long continuous recordings, released as part of the open-source VibeVoice project.
Facebook AI Research's speech recognition toolkit, whose development moved into the Flashlight machine learning library.
- WeNet
WeNet community · Self-hosted
An open-source, production-oriented end-to-end speech recognition toolkit.
- Julius
Julius development team · Self-hosted
An open-source large-vocabulary speech recognition decoder for speech researchers and developers.
- Whisper-Streaming
Charles University ÚFAL · Self-hosted
A real-time streaming transcription and translation tool built on Whisper, from Charles University's ÚFAL.