NVIDIA Parakeet
NVIDIA's family of open-weight speech-to-text models for fast automatic speech recognition.
Commenters praise it as fast, accurate on-device speech-to-text1234, often preferred over Whisper5678 and used daily for dictation through apps like Handy and Hex.9101112 The most common caveat is weaker multilingual handling, such as misdetected languages and missed non-English terms131415, and a few users still find Whisper more accurate for their needs.16
Why people recommend it
Speed · 3
Commenters call it near-instant, with one reporting 282x to 320x faster than real time in an optimized runtime.
“sped up Parakeet V2 and V3 to 320x and 282x faster than real time”
“near instant transcription with Parakeet V3”
Show 1 more
“was completely surprised at just how fast it was for being on-device!”
Daily dictation · 3
Many use it every day for dictation through apps like Handy, Hex and Superwhisper.
“i am using parakeet with https://handy.computer daily and it's amazing.”
“I use Parakeet V3 via this tool and it is actually quite reliable for me (in English)”
Show 1 more
“I use https://github.com/cjpais/Handy on Parakeet streaming”
Beats Whisper · 3
Several commenters say it is more accurate and faster than Whisper, with one calling it the gold standard.
“I would choose Parakeet over Whisper in almost all situations these days.”
“it's perfect with parakeet.”
Show 1 more
“Parakeet is the gold standard.”
Runs locally · 3
Users are impressed that it runs well on-device, including on older Macs and an iPhone.
“Nvidia's Parakeet v3 is good enough for my own just local transcription most of the time.”
Show 1 more
“I would choose Parakeet over Whisper in almost all situations these days.”
Talking to agents · 2
Its speed and accuracy suit dictating prompts to AI coding agents, where small errors matter less.
“dictation was solved a while back with Parakeet V3”
“near instant transcription with Parakeet V3”
Voice pipelines · 3
Commenters use it as the speech-to-text stage in local voice assistants and custom devices.
“Parakeet for STT”
Show 1 more
“offloads transcription to Parakeet running in the browser connected to the ESP32”
Caveats and dissent
Multilingual accuracy · 3
V3 can misdetect languages, cannot be forced to one language, and misses business terms in other languages.
“I speak Polish and it detects Russian.”
“I love it but it misses the business specific words when in different language.”
Show 1 more
Varies by user · 1
One heavy dictation user finds Whisper Large v3 more accurate for their needs despite Parakeet's speed.
“I keep hearing that Nvidia Parakeet models are better, but they just don't work as well for me”
Ecosystem friction · 2
It normally needs CUDA, requires ONNX on AMD, and its many implementations make embedding it in apps hard.
“Parakeet TDT v3 is half the size, way faster, and lower WER”
“Parakeet is the gold standard.”
Digit output · 1
It prints spoken numbers as digits, which one developer says breaks merging of transcription windows.
Often compared with
3 moved from Whisper to NVIDIA Parakeet
2 moved from NVIDIA Parakeet to WhisperHandy35 comments mention both
1 moved from Voxtral to NVIDIA Parakeet
Who recommends it in their own words
From the makers not counted in the ranking
“NVIDIA Parakeet running locally in the browser”
Threads where it came up
- Apple's new SpeechAnalyzer API, benchmarked against Whisper and its predecessor 13 mentions
- Show HN: Ghost Pepper – 100% local hold-to-talk speech-to-text for macOS 9 mentions
- Whistle: Speech to Text in 16.9 MB 8 mentions
- Ask HN: What's the current best local/open speech-to-speech setup? 8 mentions
- Show HN: Moonshine Open-Weights STT models – higher accuracy than WhisperLargev3 7 mentions
- Show HN: Free Alternative to Wispr Flow, Superwhisper, and Monologue 7 mentions
Also written as Parakeet V3, Parakeet v3, NVIDIA Parakeet, Parakeet TDT v3. Data: nvidia-parakeet.json. Quotes are verbatim; the verdict and theme labels are written by a model from these comments, and every clause links to its sources.
All comments 45
“transcribes with NVIDIA's Parakeet TDT 0.6B v2”
“I used nemotron 3.5 int4 and parakeet 0.6 tdt v3 int4 as the transcriber.”
“I rolled my own: podcast-dl + parakeet + Gemini.”
“Parakeet is pretty amazing, though”
“I've got a custom version of Parakeet Ultra I'm training now”
“been pretty impressed with the local performance”
“dictation was solved a while back with Parakeet V3”
“ported my little voice typing gui to use Whistle/Needle instead of Parakeet”
“parakeet for STT”
The good thing with most NNs is that you can go with fp32 almost always (I put almost because I'm not 100% sure that this covers all the cases) for almost 0 performance degradation. For Parakeet I'm 100% sure because I used it yesterday and it was using just 1GB. Why do you use GLM 5.3 tho? Unless you have an instance running "free" 24/7 you could use a local Gemma model. It should be good enough, no? Or maybe with o
“Parakeet TDT 0.6B v3 (int8 CoreML, about 470 MB) runs on-device through FluidAudio's AsrManager”
“offloads transcription to Parakeet running in the browser connected to the ESP32”
“Parakeet is great I use it everyday”
“Yes, it is much, much faster and smaller, but accuracy matters a lot”
“Parakeet is the gold standard.”
“I use https://github.com/cjpais/Handy on Parakeet streaming”
“parakeet for speed”
“ie using Parakeet”
“vanilla parakeet i use in handy”
“asr (parakeet v3), tts (omnivoice) take no more then 400-450 ms total”
“Parakeet for transcription and senior for diarization.”
“Also whisper is pretty outdated vs parakeet”
“I keep hearing that Nvidia Parakeet models are better, but they just don't work as well for me”
“I love it but it misses the business specific words when in different language.”
“I use Parakeet 3.”
“it's a local speech to text tool using NVIDIA Parakeet”
“(using Parakeet)”
“near realtime spoken word translation/transcription project using selfhosted Gemma4 and Nvidia Parakeet LLM models”
I’m a dictaphone fan. Even better than the kind shown in the article are the ones with sliding switches. I have an Olympus model. Mine has both internal memory and SD card capability. And similar to what the author aspires to, I record bits and bobs throughout the day and then use the SD card to transfer them to my Mac. From there, they automatically get added to a MacWhisper watch folder and then automatically trans
“near instant transcription with Parakeet V3”
“I speak Polish and it detects Russian.”
“using Parakeet V3”
“sped up Parakeet V2 and V3 to 320x and 282x faster than real time”
“Parakeet has great mandarin>CN text”
“Nvidia's Parakeet v3 is good enough for my own just local transcription most of the time.”
No, there are better open weight models: https://artificialanalysis.ai/speech-to-text/non-streaming Apparently MOSS-Transcribe-Diarize is quite good too as it released only a few days ago.
“It's great for me when configured to use Parakeet v3.”
I am running the MLX fork https://huggingface.co/vanch007/mlx-MOSS-Transcribe-Diarize Generally labs don't release MLX or ANE versions and we must rely on finding someone who's converted it Parakeet is not multilingual so not directly comparable Where do you see 16GB? MOSS is smaller than Parakeet at 1.82GB
“I use Parakeet V3 via this tool and it is actually quite reliable for me (in English)”
“i am using parakeet with https://handy.computer daily and it's amazing.”
“it's perfect with parakeet.”
“I would choose Parakeet over Whisper in almost all situations these days.”
“Parakeet TDT v3 is half the size, way faster, and lower WER”
“you can use parakeet v3 which I find way better than whisper”
“was completely surprised at just how fast it was for being on-device!”