Picovoice Cheetah (on-device streaming STT)
Commercial on-device streaming STT SDK running locally on desktop, mobile, web and Raspberry Pi. Leopard is the on-device batch counterpart. Requires a Picovoice AccessKey.
Overview
Best for: Offline or privacy-critical apps (kiosks, embedded, mobile) that need streaming STT without sending audio to a cloud.
At a glance
On-device SDK, audio never leaves the device. Free tier is non-commercial only. Price is the Foundation Plan ($6,000/yr for 25K min/month) fully used; unused minutes raise the rate. 8 languages, more on Enterprise.
16-bit PCM frames of Cheetah's frame_length (16 kHz mono in Picovoice SDKs; not restated on the doc page read).
n/a
English, French, German, Italian, Japanese, Korean, Spanish, Portuguese; more for Enterprise.
Runs locally; no network round trip. No numeric claim captured.
On-device; audio never leaves the device.
Audio processed on device (privacy by design).
Runs on CPU: Linux x86_64, macOS (x86_64, arm64), Windows (x86_64, arm64), Android, iOS, Web (WASM), Raspberry Pi 3/4/5.
Proprietary SDK; AccessKey required.
Features
- partial transcripts
- endpoint detection (is_endpoint flag in SDK)
- automatic punctuation (option)
- custom vocabulary with optional IPA
- keyword boosting
Pricing
| What | Price | Unit |
|---|---|---|
| Free tier | $0 | |
| Foundation Plan (startups) | $6,000 | per year |
| Enterprise | custom |
Foundation Plan if fully used: $6,000 / (25,000 min x 12) = $0.02/min. Unused minutes raise the effective rate.
Free tier: Non-commercial free tier; commercial free trial grants Foundation Plan rights.
Source: picovoice.ai
Setup
- Sign up at Picovoice Console and copy your AccessKey.
- pip install pvcheetah pvrecorder (SDKs also for Android, iOS, Web, C, .NET, etc.).
- Feed frames of cheetah.frame_length samples; print partials; call flush() at endpoints.
Endpoint
n/a (local library)Authentication
AccessKey passed to create(); validated against your account.
Quick start python
import os
import pvcheetah, pvrecorder # pip install pvcheetah pvrecorder
cheetah = pvcheetah.create(access_key=os.environ["PV_ACCESS_KEY"],
enable_automatic_punctuation=True)
rec = pvrecorder.PvRecorder(frame_length=cheetah.frame_length)
rec.start()
try:
while True:
partial, is_endpoint = cheetah.process(rec.read())
print(partial, end="", flush=True)
if is_endpoint:
print(cheetah.flush())
except KeyboardInterrupt:
pass
finally:
rec.stop(); rec.delete(); cheetah.delete()Written from the current docs. Check the vendor's SDK version before you ship.
Warnings
Free tier is non-commercial
Picovoice describes the free tier for personal non-commercial projects; shipping a product needs a paid plan (Foundation $6,000/yr for eligible startups, otherwise Enterprise).
AccessKey still meters usage
Even though inference is local, the SDK requires an AccessKey that is checked against account limits; plan for that dependency and key protection in apps.
Limited languages
Eight languages out of the box; others only through Enterprise.
Accuracy below cloud leaders
On-device models trade accuracy for privacy and cost; benchmark against your audio before choosing.
Plus 12 warnings that apply to all speech-to-text, live APIs. See category warnings.
Limits
- Usage metered against account limits via AccessKey.
- Commercial use requires a paid plan.
Models and products
| Name | Status |
|---|---|
| Cheetah | GA |
| Leopard | GA |
Docs and sources
Docs
Sources used
- picovoice.ai/docs/cheetah/
- picovoice.ai/checkout
- picovoice.ai/blog/introducing-picovoices-free-tier
- picovoice.ai/docs/faq/general
Current free-tier status in 2026 (pricing page did not render), exact SDK signature details (create(enable_automatic_punctuation), process() returning (partial, is_endpoint)) are from prior SDK versions.