Fast multilingual
Try Piper free — unlimited, in your browser
Piper is a family of small VITS voices built for the Rhasspy voice assistant. Each voice is a separate 27–115 MB download that synthesizes many times faster than real time on any CPU, across ~35 languages. Quality is clearly synthetic but crisp and highly intelligible — ideal for low-end devices and long documents.
Specifications
| Developer | Rhasspy / Michael Hansen |
|---|---|
| Parameters | 20 million |
| Download size | 27-115 MB per voice (fp16 per-voice) |
| Output | 22.1 kHz mono |
| Voices | 900 |
| Languages | English (US), English (UK), Spanish, French, German, Italian, Portuguese (BR), Russian, Chinese, Dutch, Polish, Turkish, Ukrainian, Czech, Danish, Norwegian, Swedish, Finnish, Greek, Hungarian, Romanian, Vietnamese, Arabic, Catalan |
| Voice cloning | No |
| Streaming | No |
| License | MIT (core) — Core code is MIT; individual voice models carry their own dataset licenses — check each voice card before commercial use. |
| Speed | ≈ 9× faster than real-time on CPU — no GPU needed at all. |
| Runtime | WebAssembly (any CPU) |
Source: rhasspy/piper-voices
Strengths
- +Fastest CPU synthesis of any model here (RTF ≈ 0.11)
- +Enormous voice/language coverage (2,600+ speakers)
- +Tiny per-voice downloads
Limitations
- −Noticeably synthetic prosody vs Kokoro
- −One speaker per voice file; no cloning
Voices (48)
Compare
Piper — questions
Is Piper free to use?
Yes. The weights are MIT (core) and they run on your device. Core code is MIT; individual voice models carry their own dataset licenses — check each voice card before commercial use.
What hardware does Piper need?
Any modern browser — it runs on CPU via WebAssembly, no GPU needed. The download is 27-115 MB per voice, cached after first use.
Is my text private when using Piper here?
Yes. The model executes inside your browser tab — the text you synthesize is never transmitted to a server.
Can I use Piper output commercially?
Review the current model, voice, and source-dataset terms before commercial use. The catalog lists MIT (core) for this model and notes: Core code is MIT; individual voice models carry their own dataset licenses — check each voice card before commercial use.