vsref
NVIDIA Parakeet / Riva logo

NVIDIA Parakeet / Riva Review

Open-weights, GPU-accelerated self-hosted STT stack

Among the 43 speech-to-text tools we track, NVIDIA Parakeet / Riva has the 31st-widest language coverage.

See pricing

Facts verified Jul 20, 2026Try NVIDIA Parakeet / Riva

If you sign up through links on this page, vsref may earn a commission; programs exist on both sides of most comparisons, and commissions never change verdicts. How we make money

What we know about NVIDIA Parakeet / Riva

This is our verified profile of NVIDIA Parakeet / Riva, a speech-to-text apis platform - open-weights, GPU-accelerated self-hosted STT stack. Every fact about NVIDIA Parakeet / Riva below carries the source it came from and the day we checked it.

NVIDIA Parakeet / Riva does not publish a public per-minute rate we have been able to verify, so the price figure here stays blank until we can confirm one. We would rather show nothing than a guessed number; if you have a current NVIDIA Parakeet / Riva quote, it is the fastest way for us to close that gap.

On capabilities, NVIDIA Parakeet / Riva covers language auto-detection, word-level timestamps, custom vocabulary / keyterm boosting, speech translation, and self-host / on-prem option, and does not offer websocket streaming api. Each of those is verified against NVIDIA Parakeet / Riva's own docs or dashboard, not marketing copy.

Among the 43 speech-to-text tools in our matrix, NVIDIA Parakeet / Riva leads with the 31st-widest language coverage; the fact sheet below has the raw numbers behind that placement.

In total we track 21 verified facts for NVIDIA Parakeet / Riva today, each linking the primary source it came from so you can check our work - and vendor claims we have not measured ourselves are labeled as such on the NVIDIA Parakeet / Riva fact sheet below.

Reviewed by vsref Editorialfacts verified Jul 20, 2026Methodology →

Fact sheet

Pricing
Pricing facts
Pricing modelhybridJul 20
Free tier quotaFree hosted trial APIs on build.nvidia.comJul 20
Capabilities
Capabilities facts
WER (third-party benchmark)6.43 % WERJul 20
WER (vendor-claimed)~6.34 % WERJul 20
Languages supported25Jul 20
Language auto-detection✓ YesJul 20
Speaker diarization✓ IncludedJul 20
Word-level timestamps✓ YesJul 20
Custom vocabulary / keyterm boosting✓ YesJul 20
Speech translation✓ YesJul 20
Compliance & trust
Compliance & trust facts
Self-host / on-prem option✓ YesJul 20
Model weights licenseCC-BY-4.0Jul 20
Build experience
Build experience facts
Official SDKsPython, Go clients; gRPC protos; CLI clientsJul 20
Max file size / duration24 min full attention; up to 3 hr with local attentionJul 20
Supported audio formatsWAV, FLAC (16 kHz mono); Opus streams in RivaJul 20
Websocket streaming API✗ NoJul 20
Commercial
Commercial facts
Model size (parameters)Parakeet TDT 0.6B (600M); Canary 1B v2 (978M)Jul 20
Hardware to self-hostNVIDIA GPU (T4 to H100); min 2GB RAMJul 20
Hosted API availableFree trial NIM endpoints on build.nvidia.comJul 20
Project maintenance statusActive - NeMo v2.7.3 released 2026-04-23Jul 20
GitHub stars17,800Jul 20

Considering a switch? Best NVIDIA Parakeet / Riva alternatives →

NVIDIA Parakeet / Riva head-to-head

Call CenterswonDeveloperslostDictationwonMedicallostMeetingswonSelf-HostedwonVoice Agentstie
DeveloperslostDictationtieMedicallostSelf-HostedwonVoice Agentslost