senstella /parakeet-mlx

An implementation of the Nvidia's Parakeet models for Apple Silicon using MLX.

★ 996⑂ 62👁 18⑃ 17● 40▤ 0
PythonApache-2.0Created 2025-05-03
+0
Daily
996
stars
62
forks
18
watchers
17
prs
40
issues
0
releases

README

Parakeet MLX

An implementation of the Parakeet models - Nvidia's ASR(Automatic Speech Recognition) models - for Apple Silicon using MLX.

 Installation

 [!NOTE]
 Make sure you have ffmpeg installed on your system first, otherwise CLI won't work properly.

Using uv - recommended way:

bash
uv add parakeet-mlx -U


Or, for the CLI:

bash
uv tool install parakeet-mlx -U


Using pip:

bash
pip install parakeet-mlx -U


 CLI Quick Start

bash
parakeet-mlx  [OPTIONS]


 Arguments

- audio_files: One or more audio files to transcribe (WAV, MP3, etc.)

 Options

- --model (default: mlx-community/parakeet-tdt-0.6b-v3, env: PARAKEET_MODEL)
  - Hugging Face repository of the model to use
  - https://huggingface.co/collections/mlx-community/parakeet

- --output-dir (default: current directory)
  - Directory to save transcription outputs

- --output-format (default: srt, env: PARAKEET_OUTPUT_FORMAT)
  - Output format (txt/srt/vtt/json/all)

- --output-template (default: {filename}, env: PARAKEET_OUTPUT_TEMPLATE)
  - Template for output filenames, {parent}, {filename}, {index}, {date} is supported.

- --highlight-words (default: False)
  - Enable word-level timestamps in SRT/VTT outputs

- --verbose / -v (default: False)
  - Print detailed progress information

- --decoding (default: greedy, env: PARAKEET_DECODING)
  - Decoding method to use (greedy or beam)
  - beam is only available at TDT/RNNT models for now

- --chunk-duration (default: 120 seconds, env: PARAKEET_CHUNK_DURATION)
  - Chunking duration in seconds for long audio, 0 to disable chunking

- --overlap-duration (default: 15 seconds, env: PARAKEET_OVERLAP_DURATION)
  - Overlap duration in seconds if using chunking

- --beam-size (default: 5, env: PARAKEET_BEAM_SIZE)
  - Beam size (only used while beam decoding)

- --length-penalty (default: 0.013, env: PARAKEET_LENGTH_PENALTY)
  - Length penalty in beam. 0.0 to disable (only used while beam decoding)

- --patience (default: 3.5, env: PARAKEET_PATIENCE)
  - Patience in beam. 1.0 to disable (only used while beam decoding)

- --duration-reward (default: 0.67, env: PARAKEET_DURATION_REWARD)
  - From 0.0 to 1.0,  0.5 to favor duration logprobs more. (only used while beam decoding in TDT models)

- --max-words (default: None, env: PARAKEET_MAX_WORDS)
  - Max words per sentence

- --silence-gap (default: None, env: PARAKEET_SILENCE_GAP)
  - Split sentence if it exceeds silence gap provided (seconds)

- --max-duration (default: None, env: PARAKEET_MAX_DURATION)
  - Max sentence duration (seconds)

- --fp32 / --bf16 (default: bf16, env: PARAKEET_FP32 - boolean)
  - Determine the precision to use

- --full-attention / --local-attention (default: full-attention, env: PARAKEET_LOCAL_ATTENTION - boolean)
  - Use full attention or local attention (Local attention reduces intermediate memory usage)
  - Expected usage case is for long audio transcribing without chunking

- --local-attention-context-size (default: 256, env: PARAKEET_LOCAL_ATTENTION_CTX)
  - Local

Top issues

Discussions

Snapshot: 2026-11-16T09:23:14Z · Data license Apache-2.0 · JSON