Parakeet English-Hebrew code-switching model, now in MLX for Apple Silicon

An MLX port of v3 of my English-Hebrew code-switching Parakeet fine-tune, so it runs natively on Apple Silicon Macs via parakeet-mlx or mlx-audio. Same 4.4% WER as the NeMo original on the check set.

Daniel Rosehill Update Evaluations 18 October 2026 (7 Cheshvan 5787) 1 min read

v3 of my English-Hebrew code-switching Parakeet fine-tune now has an MLX build: Parakeet-TDT-0.6B-v2-English-Hebrew-Codeswitching-mlx. It joins the NeMo, ONNX, sherpa-onnx and GGUF versions, so Mac users can run it natively on Apple Silicon.

Hugging Face Model danielrosehill/Parakeet-TDT-0.6B-v2-English-Hebrew-Codeswitching-mlx MLX weights for Apple Silicon of danielrosehill/Parakeet-TDT-0.6B-v2-English-Hebrew-Codeswitching, a fine-tune of nvidia/parakeet-tdt-0.6b-v2 for English speech with Hebrew words dropped in ("I have to go to Bituach… Automatic speech recognitionCC-BY-4.0Updated Oct 2026 View model on Hugging Face

Use it

With parakeet-mlx:

GitHub Repository senstella/parakeet-mlx An implementation of the Nvidia's Parakeet models for Apple Silicon using MLX. ★ 987PythonApache-2.0Updated Oct 2026 View on GitHub

pip install parakeet-mlx
parakeet-mlx audio.wav --model danielrosehill/Parakeet-TDT-0.6B-v2-English-Hebrew-Codeswitching-mlx

It also loads in mlx-audio.

GitHub Repository Blaizzy/mlx-audio A text-to-speech (TTS), speech-to-text (STT) and speech-to-speech (STS) library built on Apple's MLX framework, providing efficient speech analysis on Apple Silicon. ★ 8.1kPythonMITUpdated Oct 2026 apple-siliconaudio-processingmlxmultimodalspeech-recognition View on GitHub

How it was checked

This is a direct weight remap, with no retraining or quantisation. It uses the same conversion as the stock mlx-community Parakeet repos (senstella's script). Compared with the NeMo PyTorch output on held-out code-switching clips, 7 of 10 transcripts were identical. WER against the reference was 4.4% for both. The differences are small wording variations, not degradation.

Weights are fp32 (parakeet-mlx casts to bf16 at load). The repo tracks the main model and is replaced whenever a new version is released.