Models

The two speech models pktito uses, where to download them, and how to use a model file you already have.

The two models

The multilingual model covers Bulgarian, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, German, Greek, Hungarian, Italian, Latvian, Lithuanian, Maltese, Polish, Portuguese, Romanian, Russian, Slovak, Slovenian, Spanish, Swedish and Ukrainian. Its model card gives the accuracy for each.

Switch models from the model menu on the start screen.

Downloads

pktito checks each download against the size and SHA-256 below. A file’s address holds the start of its SHA-256, so it never changes.

parakeet-tdt-0.6b-v2

• Size: 624,480,196 bytes
• SHA-256: 8bfa6750b83cd7c447bad82057c7a12473563b72b3d35487fcca6da0786d83c7
• https://models.pktito.net/8bfa6750b83c/parakeet-tdt-0.6b-v2.pkf

parakeet-tdt-0.6b-v3

• Size: 647,552,644 bytes
• SHA-256: 35a59c9a433d4da2289640db93e73c42478b576e49a293090ef61b4ae595152a
• https://models.pktito.net/35a59c9a433d/parakeet-tdt-0.6b-v3.pkf

To check a file by hand: shasum -a 256 on macOS, sha256sum on Linux, Get-FileHash in Windows PowerShell.

Using a model file you already have

• In the app: choose Open a model file… from the model menu on the start screen, or I already have a model file… on the first run. Double-clicking a .pkf also opens pktito with it in use.
• In pktito-cli: -m takes the file’s path, as in pktito-cli -m /path/to/parakeet-tdt-0.6b-v2.pkf talk.mp4. See pktito-cli.

If pktito can’t load the file, the model in use stays as it was; Troubleshooting explains the message.

Attribution

The models are NVIDIA’s parakeet-tdt-0.6b-v2 and parakeet-tdt-0.6b-v3, under CC BY 4.0, modified by pktito (partly quantized to int8). Each file also holds Silero VAD (MIT), which finds the pauses in long recordings. pktito is not affiliated with or endorsed by NVIDIA.