Piper TTS on Mac: Piper vs Murmur and Local Options

Direct answer: Piper TTS is a fast, local, open source text to speech engine that you run from a terminal or call from code, not a Mac app with a window and buttons. It is a great fit if you are comfortable with Python, a command line, and choosing voice files by name, and it covers a long list of languages with very small models. If what you actually want is a Mac app that handles voice cloning, voice design, multi-speaker projects, and a one-click export, a finished studio app such as Murmur is the more direct path. The two are not really rivals: Piper is a building block, and Murmur is the assembled desk.

Disclosure: we make Murmur, a $49 one-time Mac app that runs speech models locally, so we have a stake in this comparison. Every claim about Piper below comes from the project's own repositories and documentation, linked in the sources at the end and checked on October 14, 2026. Piper's ownership and license changed recently, so treat this page as a starting point and read the current files yourself before you commit to anything commercial.

What Piper TTS actually is

Piper is a neural text to speech engine built to run locally on modest hardware. The original repository, rhasspy/piper, was archived by its owner on October 6, 2025, and its README now points to a new home under the Open Home Foundation: OHF-Voice/piper1-gpl. That current engine is released under GPL-3.0 and, in the project's own words, embeds espeak-ng for phonemization. Concretely, Piper first converts your text into phonetic units with espeak-ng and then a small neural model turns those units into audio. That design keeps the models tiny and quick, which is exactly why it turns up inside things like Home Assistant.

The project ships a command-line interface, a Python API, a C and C++ library called libpiper, and a small HTTP web server. Voices are VITS models exported to onnxruntime, and each voice is delivered as two files: a .onnx model and a matching .onnx.json config. The maintainers publish voice files on Hugging Face and provide a samples page so you can hear a voice before you download it. It is a toolkit for developers and tinkerers, and it is honest about being one.

Voice sizes and quality levels

Piper voices are trained at one of four quality levels, and the project documents the tradeoff plainly. Lower levels are faster and smaller; higher levels use more parameters and produce wider-band audio.

• x_low: 16 kHz audio, 5 to 7 million parameters.
• low: 16 kHz audio, 15 to 20 million parameters.
• medium: 22.05 kHz audio, 15 to 20 million parameters.
• high: 22.05 kHz audio, 28 to 32 million parameters.

Some voices bundle multiple speakers in a single model, which lets you switch speakers quickly. The project notes that the quality of any individual speaker in a multi-speaker model can be lower than a dedicated single-speaker voice. This is useful context: in Piper, picking a voice is a decision about size, sample rate, and speaker, and you make that decision by reading a file name and a model card, not by opening a voice browser.

How to run Piper TTS on a Mac today

Piper is cross-platform, and the fastest way onto a Mac is the Python package. The steps below follow the project's own command-line documentation, which was checked on October 14, 2026.

1. Install Python 3 and confirm pip works in your terminal.
2. Install the engine with pip install piper-tts.
3. List and download a voice, for example python3 -m piper.download_voices en_US-lessac-medium, which saves the two voice files to your current directory.
4. Generate audio with python3 -m piper -m en_US-lessac-medium -f test.wav -- 'This is a test.' to write a wav file, or omit -f to play through your speakers if ffplay is installed.
5. Point the tool at another folder with --data-dir if you keep voices somewhere else.
6. Open the .onnx.json config and the MODEL_CARD for the voice before you ship anything, because the license lives there.

One practical wrinkle the docs call out: the command-line tool loads the model on every run, so it can feel slow if you are generating one phrase at a time. For repeated use, the project recommends its HTTP web server instead. The docs also list a --cuda flag for GPU acceleration, but that flag requires the onnxruntime-gpu package and targets Nvidia hardware, so on a Mac you are effectively running on the CPU. The prebuilt binary releases are for Linux and Raspberry Pi, not macOS, which is another reason the pip route is the natural one here.

Where Piper TTS is genuinely the better choice

• You want to embed speech in your own program and you are comfortable writing Python, C, or C++.
• You need speech on Windows, Linux, or a Raspberry Pi, not just on a Mac.
• You are building a service or a home automation setup and want a small, self-hosted HTTP endpoint.
• You want the widest language reach from tiny models, since Piper publishes voices across 40 or more languages and regional variants.
• You want full control over phonemes, including injecting raw espeak-ng phonemes with a [[ ... ]] block to force a pronunciation.
• You prefer to pay nothing and accept that you are assembling the workflow yourself.

That last point is not a small one. Piper's license for the engine is free and open source, and the engine runs fine on a laptop with no account and no meter. If your job is to write code that produces speech, Piper is a sensible default and there is no reason to pay for an app.

Where a Mac app such as Murmur fits

The gap opens when the hard part is no longer the engine but the project around it. Murmur is a local-first macOS app for Apple Silicon Macs running macOS 15 or later. It costs $49 one time, has a free plan and a 3-day Pro trial, and carries a 7-day refund policy. The bundled Kokoro model provides preset voices and covers a set of languages including English, Spanish, Japanese, Chinese, French, Hindi, Italian, and Portuguese; the in-app model list is the source of truth for what is installed on your machine. Beyond Kokoro, optional models add reference voice cloning and a Voice Design mode that creates a voice from a written description.

The honest summary of that table is that Piper wins on reach and on price, and Murmur wins on the last mile. If your output is a script in a repository, the app is overhead. If your output is a finished file with a corrected pronunciation, a chosen voice, and a clean export, the terminal is overhead. The guide to local TTS on a Mac without Python and the local voice studio versus DIY open source comparison cover that last-mile gap in more detail.

A five minute decision checklist

1. Write down the deliverable. If it is code that emits audio, choose Piper. If it is finished audio for a video, book, or course, lean toward an app.
2. Check your platform. Need Windows or Linux? Piper, because Murmur is macOS only.
3. Check your talent gap. Uncomfortable with Python and a terminal? Piper's setup is the real cost, not the download.
4. Check the voice you need. If a Piper voice already sounds right for your project, you may not need anything else.
5. Check cloning or design. If you need to reproduce a specific voice or invent one from a description, Piper is not the tool for that out of the box.
6. Check the license of the exact voice file, not just the engine repository.
7. Only then compare price and time. Piper is free but costs you setup and maintenance; a one-time app trades money for time.

If you are still surveying options, the roundup of best local TTS models and the Kokoro TTS guide for Mac show what the model layer looks like on Apple Silicon, and the one-time app versus subscription breakdown handles the pricing question for recurring work.

Licensing: read the model card, not just the repo

This is the part most Piper comparisons skip, and it matters. The original rhasspy/piper repository carried an MIT license, but the current engine at OHF-Voice/piper1-gpl is GPL-3.0. Those are different obligations, and if you embed the engine in a product you need to understand what GPL-3.0 requires of your code.

The voices are a separate question again. Piper's own documentation states that the .onnx.json and MODEL_CARD files carry the licensing information for each voice, that Piper is intended for personal use and text to speech research only, and that while the project does not impose additional restrictions on voice models, some individual voices have restrictive licenses and should be reviewed carefully. In plain terms: a free engine does not mean every voice is free to use in commercial work. Check the specific model card of the specific voice you plan to ship with.

• Engine license: check the current repository (GPL-3.0 at the time of writing), because the project moved and the license changed.
• Voice license: read the MODEL_CARD and the .onnx.json for the exact voice file you download.
• Training data terms: some voices are derived from datasets with their own conditions.
• When in doubt, choose a voice whose card grants the use you need, or generate your own sample for a clone, rather than assuming the engine's license covers the output.

Murmur is a commercial app, and the practical point of that is that you are paying for a defined product and its model list rather than assembling licenses yourself. That is not automatically the cheaper or the safer choice in every case, but it removes a category of homework for creators who would rather not read model cards.

Frequently asked questions