Best Open-Source Text-to-Speech Engines 2026: 6 Options Compared
Best open-source Text-to-Speech Engines compared: Piper, Kokoro, Chatterbox, eSpeak NG and 2 more, with licenses, platforms and prices from official pages checked October 24, 2026.
•3 min read•Voice AI & Transcription
Quick comparison
| Tool | Price | Platforms | Best for | Tested |
|---|---|---|---|---|
| Piper | Free (Free local neural text-to-speech engine) | Self-hosted (Python package) | Offline reading and home automation | Docs only |
| Kokoro | Free (Free open-weight 82M-parameter model you run yourself) | Self-hosted (Python library) | Natural voices on small hardware | Docs only |
| Chatterbox | Free (Free open-source models (Nano 110M runs on CPU) with voice cloning from a short reference clip) | Self-hosted (Python library) | Cloning a voice you have rights to | Docs only |
| eSpeak NG | Free (Free and open source; no paid tier stated on the source page) | Self-hosted (C) | Tiny footprint and many languages | Docs only |
| Coqui TTS (idiap fork) | Free (Free open-source library; XTTS clones a voice from a speaker reference file) | Self-hosted (Python) | Experimenting with many models | Docs only |
| Bark | Free (Free and open source; no paid tier stated on the source page) | Self-hosted (Python) | Expressive, experimental audio | Docs only |
Independent comparison. Paid placements are labelled separately; editorial links earn no affiliate commissions. Learn more.
Good fit if
- You want offline narration that runs on a small machine (Piper is the pick)
- You want the most natural voice for the size (Kokoro is the pick)
- You need a voice clone you have permission for (Chatterbox is the pick)
- You need many languages in a tiny package (eSpeak NG is the pick)
- You want to try several models in one toolkit (Coqui TTS (idiap fork) is the pick)
Not ideal if
- You cannot host, patch and back up a server yourself and want a vendor to run it
Open-source text to speech splits into two eras. The older engines, such as eSpeak NG, are tiny and fast, support many languages, and sound clearly robotic; they remain useful for accessibility, embedded devices and quick prototypes. The newer neural models sound far more natural but need more compute. Piper is a local neural engine that runs quickly on modest hardware, making it a favorite for home automation and offline reading. Kokoro is a small open-weight model with strong quality for its size, Apache-licensed. Chatterbox, MIT-licensed, adds voice cloning from a short reference clip, which brings consent and misuse questions you should take seriously. Coqui TTS is a toolkit with many models and a cloning model, and Bark generates expressive speech and sound effects, though with less predictable results. Licensing deserves a careful look, because code and model weights can carry different terms, so confirm the voice or checkpoint you pick allows your use. The trade-off is polish against control: paid voice services give you a studio-grade voice with one click and legal clarity, while these tools give you privacy and no per-character fee but ask you to install Python, manage models and tune output. Listen to a paragraph of your own text in two engines before deciding.
Piper
GPL-3.0 local neural text-to-speech engine that installs with pip. Fast on modest hardware; voices come from the repository docs and have no cloning.
Price: Free (Free local neural text-to-speech engine) (Piper pricing, checked 2026-10-22).
Platforms and license: Self-hosted (Python package). GPL-3.0 (open source) (license source).
Project activity: 6.1k GitHub stars · latest release v1.8.0 (Sep 4, 2026) · last commit Oct 25, 2026
Best for: Offline reading and home automation.
Kokoro
Apache-2.0 open-weight 82M-parameter model run on your own machine. Small and natural sounding for its size, with a fixed set of voices.
Price: Free (Free open-weight 82M-parameter model you run yourself) (Kokoro pricing, checked 2026-10-22).
Platforms and license: Self-hosted (Python library). Apache-2.0 (open source) (license source).
Project activity: 9.7k GitHub stars · No commits since Aug 6, 2025
Best for: Natural voices on small hardware.
Worth knowing: Inference library with several languages via the misaki grapheme-to-phoneme library; no hosted service from the project.
Chatterbox
MIT-licensed models with voice cloning from a short reference clip; the README says each file carries a watermark. Cloning needs consent from the speaker.
Price: Free (Free open-source models (Nano 110M runs on CPU) with voice cloning from a short reference clip) (Chatterbox pricing, checked 2026-10-22).
Platforms and license: Self-hosted (Python library). MIT (open source) (license source).
Project activity: 27.3k GitHub stars · latest release v0.1.2 (Jun 13, 2025) · last commit Jul 21, 2026
Best for: Cloning a voice you have rights to.
Worth knowing: Every generated file carries a Perth watermark per the README. Multilingual V3 covers 23+ languages.
eSpeak NG
GPL-3.0 compact synthesizer covering many languages. Clearly robotic, but tiny, fast and dependable for accessibility and devices.
Price: Free (Free and open source; no paid tier stated on the source page) (eSpeak NG pricing, checked 2026-10-24).
Platforms and license: Self-hosted (C). GPL-3.0 (open source) (license source).
Project activity: 7k GitHub stars · latest release 1.52.0 (Dec 12, 2024) · last commit Sep 22, 2026
Best for: Tiny footprint and many languages.
Coqui TTS (idiap fork)
MPL-2.0 toolkit with many models, including XTTS, which clones a voice from a reference file. Check each model's own license before commercial use.
Price: Free (Free open-source library; XTTS clones a voice from a speaker reference file) (Coqui TTS (idiap fork) pricing, checked 2026-10-23).
Platforms and license: Self-hosted (Python). MPL-2.0 (open source) (license source).
Project activity: 2.3k GitHub stars · latest release v0.28.0 (Oct 19, 2026) · last commit Oct 24, 2026
Best for: Experimenting with many models.
Bark
MIT transformer text-to-audio model that can produce expressive speech. Output is less predictable than a dedicated TTS engine.
Price: Free (Free and open source; no paid tier stated on the source page) (Bark pricing, checked 2026-10-24).
Platforms and license: Self-hosted (Python). MIT (open source) (license source).
Project activity: 39.3k GitHub stars · No commits since Aug 19, 2024
Best for: Expressive, experimental audio.
Which one to pick
- You want offline narration that runs on a small machine: Piper
- You want the most natural voice for the size: Kokoro
- You need a voice clone you have permission for: Chatterbox
- You need many languages in a tiny package: eSpeak NG
- You want to try several models in one toolkit: Coqui TTS (idiap fork)
Limits of this review
Docs-only for Piper, Kokoro, Chatterbox, eSpeak NG, Coqui TTS (idiap fork), Bark: we did not install or run them, so feature claims come from vendor pages. Vendor prices and free limits change; confirm on the linked pricing page before you commit.
Related guides
More in Voice AI & Transcription
All Voice AI & Transcription alternatives →Sources
Kevin Fan
Kevin Fan is a SaaS founder. He runs Free Instead Of, which tracks free and open-source alternatives to paid software and checks each price and free-plan limit against the vendor's own pages.
Methodology
Based on official pricing, license and product pages checked between 2026-10-22 and 2026-10-24: Piper, Kokoro, Chatterbox, eSpeak NG, Coqui TTS (idiap fork), Bark. Prices, free limits and licenses are the figures shown on those pages on those dates. No tool on this page was installed or run. No accounts were created.