Parakeet v3 (TDT 0.6B): speed, accuracy and RAM on a Mac
Parakeet TDT 0.6B v3 is NVIDIA's speech-to-text model for 25 European languages, with a time for every word. It is the default transcription model in PocketWebTools for Mac. Below is what we measured on a 16 GB M1 Pro: speed, memory and accuracy on five test sets.
Which Mac runs it
- Mac memory
- 8 GB and up
- File
- 0.74 GB
One GGUF file, 8‑bit.
- Memory while transcribing
- 0.95 GB
Peak for the whole process on our test set.
- Alongside a chat model
- Works
With Qwen3.5 2B generating at the same time, transcripts were identical to a run on its own.
Parakeet v3 speed on an M1 Pro
Measured in the app's own engine on a 16 GB M1 Pro.
- Short clips, 5 to 20 s
- 60 to 72 times realtime
Across the five test sets below.
- Against Whisper Large v3 Turbo
- 4 to 8 times faster
Whisper ran the same clips at 8 to 16 times realtime.
- Picking clips from a 25.7-minute talk
- 2 min 36 s
Transcription plus Qwen3.5 9B choosing the clips.
Parakeet v3 accuracy next to Whisper
Word error rate on 170 clips with reference transcripts, 28 minutes in all, scored with the Open ASR Leaderboard normaliser. Lower is better. The clips are short, 5 to 20 seconds, and there are 25 to 40 per set.
- Read English (LibriSpeech test-other)
- 2.38%
Whisper Large v3 Turbo: 2.88%.
- Meetings (AMI)
- 7.72%
Whisper Large v3 Turbo: 14.01%.
- Earnings calls (Earnings22)
- 9.78%
Whisper Large v3 Turbo: 10.76%.
- French audiobooks
- 7.45%
Whisper Large v3 Turbo: 4.07%, the better of the two.
- German audiobooks
- 12.22%
Whisper Large v3 Turbo: 9.57%, the better of the two.
- Punctuation and capitals
- On every set
Some larger models we tested returned bare lowercase text, which a word error rate does not show.
Settings the app runs Parakeet v3 with
- File
- parakeet-tdt-0.6b-v3-Q8_0.gguf
The transcribe.cpp project's own conversion, pinned to one revision and checked against its SHA-256 after download.
- Engine
- transcribe.cpp on Metal
- Quantization
- 8‑bit (Q8_0)
- Timestamps
- Per word
What the app does with Parakeet v3
Transcribe
Drop any audio or video file this Mac can play, with no size limit, and get a transcript with timestamps. Export as text, SRT, WebVTT or JSON.
Clip
Its word times drive the captions that light up word by word, and let a clip start and end on a word instead of mid-sentence.
Edit
The video editor cuts media by deleting words from the transcript. It needs a time for every word, so it runs on Parakeet.
Where it falls short
- 25 European languages only. For anything else, and for translation into English, the app has Whisper Large v3 Turbo.
- Whisper was more accurate on our French and German sets: 4.07% against 7.45%, and 9.57% against 12.22%.
- It spells numbers out, so a transcript says ninety three percent where some models write 93%.
- The test clips are short. We have no word error rate for hour-long recordings, and none for languages other than English, French and German.
Run Parakeet v3 in PocketWebTools for Mac
One app for chat, transcription, documents, voices, photos and video, with every model on your Mac. $99 once, free lifetime updates.
Parakeet v3 questions
- Is Parakeet v3 better than Whisper?
- On our English sets, yes, and much faster. Parakeet v3 scored 2.38% on read speech, 7.72% on meetings and 9.78% on earnings calls against 2.88%, 14.01% and 10.76% for Whisper Large v3 Turbo, at 60 to 72 times realtime against 8 to 16. Whisper won on French (4.07% against 7.45%) and German (9.57% against 12.22%), and covers 99 languages to Parakeet's 25.
- How fast is Parakeet v3 on an M1 Pro?
- 60 to 72 times realtime on short clips, on a 16 GB M1 Pro with transcribe.cpp on Metal. That is 4 to 8 times faster than Whisper Large v3 Turbo on the same clips.
- How much RAM does Parakeet v3 need?
- About 0.95 GB while transcribing, measured as the peak for the whole process. The file is 0.74 GB. PocketWebTools for Mac files it under 8 GB Macs.
- Does Parakeet v3 give word timestamps?
- Yes, a time for every word. That is why the app uses it for word-by-word captions in Clip and for the transcript-based video editor.
- Which languages does Parakeet v3 support?
- 25 European languages. We measured English, French and German. It does not cover Asian languages, Arabic or Hindi.
- Can I use Parakeet v3 commercially?
- Yes. It is released under CC BY 4.0, which allows commercial use with attribution. The app ships the license text.