Skip to content
← Models
ClosedPicovoice/10 trustMetrics-checked

Picovoice Leopard

Offline speech-to-text with word timestamps, confidence scores and speaker labels — no cloud involved.

Visit official page →

About

Leopard is Picovoice's on-device (non-streaming) speech-to-text engine. Beyond the transcript it returns per-word start and end times, confidence scores, automatic punctuation and true-casing, with optional speaker diarization added in v2.0 (November 2023); v3.0 (December 2025) added GPU and multi-core support. It ships an English model by default plus French, German, Italian, Japanese, Korean, Portuguese and Spanish, and custom vocabularies can be trained in the no-code Picovoice Console. Like Cheetah, the SDKs are Apache 2.0 with a free AccessKey while the engine models are proprietary, and accuracy figures are vendor-published.

Specs

Context window

Audio-file based; offline

Pricing

Free tier with monthly transcription-hour allowance; paid plans for higher volume (see provider pricing page)

Free quota

Free tier with monthly transcription-hour allowance

API access

On-device SDKs (Python, Node.js, .NET, Java, Go, Flutter, React Native, Web); no cloud API

Weights

Closed — API only

Released

Not disclosed

Benchmarks

Vendor-published docs and benchmark repo; no independent third-party reproduction found — treat as provider-claimed.

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

meeting-transcriptionmedia-transcriptionoffline-captioningvoice-notes

Trust score — 7/10

/10 trustMetrics-checked

Scored 7/10 (metrics tier, metrics-v1). Evidence: Picovoice/leopard repo verified active (Apache-2.0 bindings; v3.0.0 changelog Dec 2025); word-metadata and language support verified in the binding READMEs. Strengths: rich offline metadata (timestamps, confidence, diarization), custom vocabularies, 8 languages. Docked for: (1) engine models proprietary and AccessKey-gated; (2) accuracy figures vendor-published with no independent check. License: SDKs Apache-2.0; models proprietary.

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
6.0/10
Maintenance · recency, releases, issue load
9.0/10
Openness · license + docs
5.0/10
Manual review · sanity read + red-flag check
8.0/10
  • · 485 stars on Picovoice/leopard; v3.0.0 released Dec 2025
  • · On-device file transcription with word timestamps, confidence, diarization (v2.0+)
  • · 8 languages: en + fr/de/it/ja/ko/pt/es
  • · Accuracy figures vendor-published; no independent reproduction

Sources