How to Clone a Voice Free, Offline, on a Mac

Short answer: You can clone a voice for free on a Mac through three honest routes, and each one has a different catch. Apple's built-in Personal Voice is free and runs on device, but it clones only your own voice and Apple licenses it for personal, non-commercial use. Open-source models such as Chatterbox and Coqui XTTS are free to download and run locally, but you manage a Python environment, and some carry a watermark or a model license you have to read. Cloud free tiers are the easiest to try today, but cloning usually sits behind a paid plan and the audio may not carry commercial rights. This guide walks each route, the steps, and the exact limit that decides whether it is truly free for your job.

Free splits into three separate costs

People search for free voice cloning because they want one of three things, and the word free means something different in each case. The first is money: no subscription, no per-character meter, no card on file. The second is rights: they intend to publish, monetize, or hand the audio to a client, so the license attached to the model matters more than the price tag. The third is effort: they are willing to pay in setup time if it removes a recurring bill.

A route can be free on one axis and expensive on another. Apple's Personal Voice is free on money and effort, but it fails on rights because your own cloned voice is limited to personal, non-commercial use. A cloud free tier is free on effort, but it fails on rights and often on cloning itself. Running an open-source model is free on money, but the effort is real and every model brings its own license. Decide which axis you actually need before you pick a tool, because the cheapest option is not always the one you can lawfully publish.

Free route 1: Apple Personal Voice on your Mac

Apple ships a genuine cloning feature called Personal Voice, and on a Mac it is completely free. According to Apple's support page, which I opened on October 14, 2026, it requires a Mac with Apple silicon and a recent macOS, and it is available only in languages Apple supports for Personal Voice. The steps are short:

1. Choose the Apple menu, open System Settings, click Accessibility in the sidebar, then click Personal Voice.
2. Click Create a Personal Voice and enter your login password. Apple notes you must have a login password set.
3. Read the sentences on screen aloud, using your Mac's microphone or an external mic. You can pause and resume the recording session at any time.
4. Wait for Apple to process the speech on device. You get a notification when the voice is ready.
5. If you want to reuse it, turn on Share across devices, and turn on Allow applications to use your Personal Voice for apps such as augmentative communication tools.

The limit is not technical, it is legal and structural. Apple states plainly that you can use Personal Voice only to create a voice that sounds like you, on device, using only your own voice, and only for your own personal, non-commercial use. It also cannot be captured: third-party apps can speak through it, but they cannot record its output. So Personal Voice is a strong free option if you need your own voice for accessibility or personal use, and it is not a way to clone a narrator for a monetized video or a client project.

Free route 2: run an open-source cloning model yourself

The second free route is to download a cloning model and run it on your Mac. These are real, capable models, not demos. Chatterbox, by Resemble AI, is a family of open-source text-to-speech models under an MIT repository license. It includes an English Turbo model and a Multilingual V3 model covering the languages its README lists, and its examples accept a device of cuda, cpu, or mps, which is the clearer Mac path. Cloning works by passing a reference clip: the example uses audio_prompt_path="your_10s_ref_clip.wav".

The older, wider-known option is Coqui TTS, a Python toolkit whose XTTS v2 model is a multilingual voice cloning model. Its documentation drives the clone from a reference file too, for example tts.tts_to_file(text=..., speaker_wav="my/cloning/audio.wav", language="en"). Both projects are honest about being developer tools: Coqui's README states it is tested on Ubuntu with Python 3.9 up to 3.11, and neither ships a Mac app, a project workspace, or an editing timeline. You install packages, point the script at a file, and manage everything in a terminal.

A practical run looks like this. Confirm a compatible Python version, install the package (for example pip install chatterbox-tts or pip install TTS), and prepare one clean reference clip of a single speaker, so the model is not averaging two voices. Then generate a short line, listen, and only then commit to a longer script. If you are new to reference recordings, the voice cloning sample recording guide covers microphone placement, room noise, and why a lightly processed sample usually beats a heavily cleaned one.

Two limits catch people who assume open source means unrestricted. First, the code license is not the model license: a permissive repository can still ship weights or dependencies with their own terms, and the toolkit license does not automatically grant commercial rights to the audio you generate. Second, some models embed a watermark. Chatterbox's README states that every audio file it generates includes Resemble AI's Perth watermark, which is designed to survive editing and compression. That matters if you plan to deliver the audio as if it were a plain recording. Read the commercial license guide for local TTS models before you use any of this in paid work.

Free route 3: cloud free tiers and what they exclude

The third route is a cloud service with a free plan. This is often the fastest way to judge whether a cloned voice clears your quality bar, because there is nothing to install. The pattern across vendors is consistent: the free tier lets you try text to speech, and the two features people want most, voice cloning and the commercial license, are the ones that move to a paid tier. As of the ElevenLabs pricing page I checked on October 14, 2026, instant voice cloning and the commercial license are paid-plan features, not free ones.

Because these tiers change often, treat any specific number as something to confirm on the vendor's own pricing page rather than trusting a blog post, including this one. Two structural points do stay true. Free cloud tiers are metered and shared across features, so heavy use of one feature leaves less for speech. And free output is uploaded, processed, and stored on someone else's servers, which is exactly what people trying to keep a voice private are trying to avoid. If privacy is the reason you are here, the private voice cloning guide walks through consent records, storage, deletion, and disclosure.

The three free routes side by side

The table lines up what each free route actually gives you, plus one paid option for comparison. Prices and terms change, so confirm the current details on each vendor's own page before you rely on them.

A short checklist to pick your route

Work down this list and stop at the first line that matches your situation. It resolves most cases in two or three steps.

1. If the voice you need to clone is your own, and the result is for personal or accessibility use, start with Apple Personal Voice. It is free, on device, and native to your Mac.
2. If you intend to publish or sell the audio, Personal Voice is out because of its personal, non-commercial limit. Move to a route that grants commercial rights.
3. If you are comfortable in a terminal and want no recurring cost, test an open-source model. Start with a clean 5 to 15 second reference clip and read the model license and watermark terms.
4. If you only want to hear whether cloned quality is good enough, use a cloud free tier for the test, but do not plan to ship from a plan that excludes commercial use or cloning.
5. If you want local cloning without managing Python, and you are on Apple silicon, compare a one-time Mac app and judge it against the same reference clip.
6. If you need Windows, Android, a browser app, live streaming, or speech to text, no free offline Mac route fits. Choose a cloud product built for that job.

A word on reference audio, because it trips up every route equally. Cloning quality depends far more on the sample than on the price of the tool. One speaker, low room noise, and enough natural variation is the target. Some models are fussier than others: in Murmur, Chatterbox Turbo needs more than five seconds of reference audio, Qwen3-TTS Base accepts from three seconds, and several engines require an exact transcript that matches the clip word for word.

When a paid local app is the honest answer

Murmur is the paid route, and it exists for a specific gap: people who want local cloning and commercial-friendly output without maintaining a Python environment. It is a local voice studio for Apple silicon Macs running macOS 15 or later. It costs $49 once, with a free plan and a 3-day Pro trial, and a 7-day refund policy. It generates speech on the machine, includes voice cloning and voice design alongside preset voices, keeps a reusable voice library, and exports WAV or MP3.

It is worth the money if you clone regularly, want a repeatable workflow, and would rather pay once than babysit dependencies. It is not the right answer if you need Windows or mobile, real-time streaming, speech to text, or free in the literal sense, because there is a free plan and a 3-day Pro trial and it is macOS only. If you want to see how the whole field lines up before deciding, the local AI voice cloning apps for Mac comparison and the best local TTS models guide are the two most useful next reads, and the general voice cloning guide covers the technique itself.

Frequently asked questions