The open-source alternatives to AssemblyAI are Whisper, whisper.cpp, WhisperX, NVIDIA's Parakeet, Moonshine, Vosk, sherpa-onnx and Speaches. AssemblyAI's product is an API that returns a transcript plus speaker labels and understanding on top; the open-source split is a transcription model and, if you want it, a separate language model for the understanding. WhisperX comes closest to the transcript with speaker labels and word timestamps. Speaches gives you an OpenAI-compatible server around Whisper. A team switches when audio has to stay inside its own network, or when a GPU costs less than the per-hour bill.
7 of 8 carry an OSI licence. 8 of 8 run on your own machines. Figures read from GitHub on October 26, 2026.
One project per need, with the reason from its own README and the one thing to know before you choose it.
BSD-2-Clause. Batched Whisper at what its README calls 70 times real time, forced alignment for word timings, and speaker labels through pyannote.
Know firstThe diarization model is CC-BY-4.0 behind a Hugging Face agreement, and the README calls its diarization far from perfect.
MIT. Streaming transcription, translation and speech generation behind OpenAI's API shape, loading Whisper, Kokoro and Piper models on demand.
Know firstThe README is short and defers to the website; the latest GitHub release is a release candidate.
MIT code and weights, six sizes, multilingual recognition plus translation to English. The model every other entry here builds on or competes with.
Know firstA model and a command line: no server, no streaming, no speaker labels of its own.
MIT. A dependency-free C and C++ port with quantised models, first-class Apple Silicon through Metal and Core ML, an HTTP server with an OpenAI-like API, and bindings for a dozen languages.
Know firstWord timestamps and speaker turns are both marked experimental.
Apache-2.0 code and CC-BY-4.0 weights. Parakeet TDT 0.6B v3 covers 25 European languages with automatic detection, punctuation, word timestamps and long audio.
Know firstThe toolkit recommends an NVIDIA GPU; the label is open weights because CC-BY-4.0 is not an OSI licence.
MIT code, MIT streaming models, built to work while the person is still talking, with models down to 1 MB and one library across Python, the browser, phones and desktops.
Know firstThirteen legacy non-English, non-streaming models stay under a non-commercial licence.
Apache-2.0. Models of about 50 MB, a streaming API, 20+ languages, speaker identification, and bindings from Python to Go, from a Raspberry Pi up to a cluster.
Know firstThe README points to the website for everything else, and the last tagged release is from April 2024.
Apache-2.0 runtime for speech to text, text to speech, speaker diarization and voice activity detection, on CPUs, NPUs, Android, iOS and WebAssembly, offline.
Know firstA lower-level tool: you pick the models and read their licences.
The facts that decide most choices. Each project's page carries the full panel.
| Project | Best for | Openness | Licence | Runs as | Own machines | Stars |
|---|---|---|---|---|---|---|
| WhisperX | Word timestamps and speaker labels | OSI | BSD-2-Clause | pip, CLI, Python library | Yes | 24,860 |
| Speaches | An OpenAI-compatible server | OSI | MIT | Docker, Server (OpenAI-compatible API) | Yes | 3,736 |
| Whisper | The reference model on a GPU | OSI | MIT | pip, CLI, Python library | Yes | 111,328 |
| whisper.cpp | A CPU, a Mac, or a phone | OSI | MIT | C/C++ library, CLI, Server (OpenAI-like API), Docker | Yes | 54,685 |
| Parakeet (NeMo Speech) | High volume in European languages on NVIDIA | Open weights | Apache-2.0 | pip, Docker, Python library | Yes | 18,619 |
| Moonshine | Live streaming on the device | OSI | See LICENSE | pip, CLI, C library, JavaScript/WASM, mobile SDKs | Yes | 11,296 |
| Vosk | Tiny models on boards and phones | OSI | Apache-2.0 | Language bindings (Python, Java, Node.js, C#, C++, Rust, Go) | Yes | 15,202 |
| sherpa-onnx | Speech inside your own app, any platform | OSI | Apache-2.0 | Libraries in 12 languages, prebuilt Android and Flutter apps, WebAssembly, websocket server | Yes | 15,528 |
Licence and star figures read from GitHub on October 26, 2026. Stars are shown as a dated fact and were not used to rank this page.
Batched Whisper transcription under BSD-2-Clause with word-level timestamps from forced alignment and speaker labels from pyannote, the closest open-source match to a hosted transcription API's output.
BSD-2-Clause. Batched Whisper at what its README calls 70 times real time, forced alignment for word timings, and speaker labels through pyannote.
Know firstThe diarization model is CC-BY-4.0 behind a Hugging Face agreement, and the README calls its diarization far from perfect.
An OpenAI-compatible server under MIT for streaming transcription, translation and text to speech, loading Whisper, Kokoro and Piper models on demand, described by its maintainers as Ollama for speech models.
MIT. Streaming transcription, translation and speech generation behind OpenAI's API shape, loading Whisper, Kokoro and Piper models on demand.
Know firstThe README is short and defers to the website; the latest GitHub release is a release candidate.
OpenAI's general-purpose speech recognition model, released with code and weights under MIT, in six sizes from 39 million to 1.55 billion parameters, with multilingual transcription, translation to English and language detection.
MIT code and weights, six sizes, multilingual recognition plus translation to English. The model every other entry here builds on or competes with.
Know firstA model and a command line: no server, no streaming, no speaker labels of its own.
A dependency-free C and C++ port of Whisper under MIT that runs on CPUs, Apple Silicon and phones, with quantised models, an HTTP server and bindings for a dozen languages.
MIT. A dependency-free C and C++ port with quantised models, first-class Apple Silicon through Metal and Core ML, an HTTP server with an OpenAI-like API, and bindings for a dozen languages.
Know firstWord timestamps and speaker turns are both marked experimental.
NVIDIA's speech models and toolkit: Apache-2.0 code, and the Parakeet TDT 0.6B v3 model under CC-BY-4.0 with 25 European languages, automatic language detection, punctuation and word timestamps.
Apache-2.0 code and CC-BY-4.0 weights. Parakeet TDT 0.6B v3 covers 25 European languages with automatic detection, punctuation, word timestamps and long audio.
Know firstThe toolkit recommends an NVIDIA GPU; the label is open weights because CC-BY-4.0 is not an OSI licence.
An on-device speech toolkit under MIT built for live streaming, with speech-to-text models trained from scratch in sizes down to 1 MB, one library across Python, the browser, phones and desktops.
MIT code, MIT streaming models, built to work while the person is still talking, with models down to 1 MB and one library across Python, the browser, phones and desktops.
Know firstThirteen legacy non-English, non-streaming models stay under a non-commercial licence.
An offline speech recognition toolkit under Apache-2.0 with streaming, models of about 50 MB, 20+ languages, a reconfigurable vocabulary and speaker identification, from small boards to clusters.
Apache-2.0. Models of about 50 MB, a streaming API, 20+ languages, speaker identification, and bindings from Python to Go, from a Raspberry Pi up to a cluster.
Know firstThe README points to the website for everything else, and the last tagged release is from April 2024.
A runtime under Apache-2.0 from the next-generation Kaldi team that runs speech to text, text to speech, speaker diarization and voice activity detection locally on CPUs, NPUs and phones, with no internet connection.
Apache-2.0 runtime for speech to text, text to speech, speaker diarization and voice activity detection, on CPUs, NPUs, Android, iOS and WebAssembly, offline.
Know firstA lower-level tool: you pick the models and read their licences.
The questions that settle it, in the order they usually come up.
WhisperX, with pyannote's gated CC-BY-4.0 model. For a desktop, Buzz, Vibe and OpenWhispr under meeting transcription label speakers too.
As a second step: transcribe with any engine here, then run the text through a language model you run or rent. No engine on this page returns both in one call.
GPU volume: WhisperX or Parakeet. Small hardware: whisper.cpp, Vosk, Moonshine or sherpa-onnx.
AssemblyAI, in the vendor's own words, is "voice AI infrastructure for developers building products that transcribe, understand, and act on speech", by its own description. assemblyai.com. The cases below are where that still wins.
Stars, licence, last commit and last release are read from the GitHub API by a script every week and carry the date they were read. Nobody types them.
What each project does is written from its own README and LICENSE, read in full on the date shown, and phrased as the project's claim. The sources are listed at the end.
That a project is faster, better or cheaper than AssemblyAI. A price for anyone. A user count. A roadmap. If a fact is not on this page, it was not in the source.
WhisperX (BSD-2-Clause) is the closest to AssemblyAI's transcript output: fast batched Whisper with word timestamps and speaker labels. For a hosted-API shape, Speaches (MIT) serves Whisper models through an OpenAI-compatible endpoint. For live audio on small hardware, Moonshine (MIT) or Vosk (Apache-2.0). For a Mac, a phone or a CPU-only server, whisper.cpp (MIT).
WhisperX, through pyannote's diarization model, which is CC-BY-4.0 and requires accepting an agreement on Hugging Face. whisper.cpp has an experimental speaker-turn mode. Vosk lists speaker identification. Buzz and Vibe, under meeting transcription, label speakers in a desktop app.
Yes, as a second step: transcribe with any engine here, then send the text to a language model you run or rent. Meetily and Vibe under meeting transcription do this in one app with Ollama for a local model. There is no single open-source engine on this page that returns transcript and summary in one call.
WhisperX's README claims 70 times real time with Whisper large-v2 on batched inference, and NVIDIA's Parakeet TDT is described as built for high-throughput transcription and handles up to 24 minutes of audio with full attention on an A100. Both figures are the projects' own.
Whisper is multilingual and also translates to English, with accuracy its README says varies widely by language. Parakeet TDT v3 covers 25 European languages with automatic detection. Vosk lists 20+ languages. Moonshine's streaming models are MIT in every language; its thirteen legacy non-English non-streaming models are under a non-commercial licence. WhisperX alignment ships for English, French, German, Spanish and Italian by default with more on Hugging Face.
No. Speaches and whisper.cpp's server are drop-ins for the OpenAI transcription API shape, so code written against OpenAI works with a URL change. Calls written against AssemblyAI's SDK have to be rewritten against whichever engine you pick.
Reveneau sets up the replacement on your own servers, connects it to your product, and builds the software around it.