NVIDIA Parakeet v2 · NVIDIA
NVIDIA Parakeet TDT v2 is the most accurate transcription model for English according to the LibriSpeech and TED-LIUM benchmarks. Faster than Whisper.
About NVIDIA Parakeet v2
NVIDIA Parakeet TDT v2 is the most accurate transcription model for English according to the LibriSpeech and TED-LIUM benchmarks. Faster than Whisper.
Where it shines: Better accuracy in English · Faster than Whisper · NVIDIA-optimized.
How to use NVIDIA Parakeet v2
-
1
Paste the audio URL
Paste a direct link to an MP3, WAV, MP4 or any audio/video hosted publicly. Supports several hours.
-
2
Generate
Click transcribe. The model processes the audio in 30–90 seconds depending on length.
-
3
Download or share
Download the result or share the direct link. No watermark, ready to use.
Frequently asked questions
How much does it cost to use NVIDIA Parakeet v2?
NVIDIA Parakeet v2 is a Pro model: it costs 144 tokens per use (~$0.72 real cost for us). You need a Pro plan ($9/month → 15,000 tokens) or a one-shot pack. If you already have tokens in the free account, you can also spend them directly.
How many uses of NVIDIA Parakeet v2 are included in the Pro plan?
Pro ($9/month) gives you 15,000 recurring tokens, at 144 tokens per use of NVIDIA Parakeet v2, that's ~104 full uses per cycle. If you run out, one-shot packs (5,000 / 25,000 / 80,000 tokens) add to the balance without expiring before one year.
What makes NVIDIA Parakeet v2 special?
NVIDIA fine-tunes its models for fast inference on its own optimized hardware — good at technical questions and reasoning, its specific strengths are: better English accuracy, faster than whisper, and nvidia-optimized.
How long does NVIDIA Parakeet v2 take to transcribe?
NVIDIA Parakeet v2 is one of the fastest models: typically ~1/10 of the audio time per generation, but noise-heavy or multilingual audio may take a little longer.
How do I use NVIDIA Parakeet v2 on ia.gratis?
You can use NVIDIA Parakeet v2 from /audio/transcribe/ by uploading the file. The REST API accepts multipart uploads in POST /api/tools/stt/. Quick summary: NVIDIA sTT. Leading English accuracy. The internal model identifier is `parakeet` — useful when integrating via API.