Skip to content
Audio

Fleet-Clean-0.36B

remove filler words from speech transcripts

2 min readupdated

This model combines two open components totaling 363M parameters to turn raw speech into clean text. The first stage (Fleet-2) handles speech-to-text, while the second (a word-level tagger built on laya-multilingual) strips out filler words, repetitions, and self-corrections, and adds punctuation. Crucially, the tagger never generates new words—it only decides whether to keep or drop existing ones. Supporting 8 languages, it is built for developers working on voice assistants and transcription systems.

Key features

  • Two-stage architecture: pairs the base Fleet-2 speech recognition model with a language tagger.
  • Multilingual support: works across English, German, French, Spanish, Italian, Dutch, Polish, and Korean.
  • Hallucination-free editing: never invents facts or adds extra text—it only strips verbal clutter and fixes punctuation.
  • Flexible tuning: lets you adjust the word-removal threshold and punctuation modes.

What you can do

  • Transcribe audio: clean up recorded conversations or podcasts by removing stutters and fillers like "uh" or "kind of."
  • Process raw text: run existing text strings through the script to strip out speech disfluencies.
  • Integrate into pipelines: pair it with speech recognition tools to get clean, ready-to-use text output.

Example

python fleet_clean.py clip.wav                  # clean transcript
python fleet_clean.py --json clip.wav           # raw, clean and the per-word tags
python fleet_clean.py --text "so I I think, you know, we should go"
python fleet_clean.py --punct full clip.wav    # also re-punctuate words away from deletions
Read moreDetails
Benchmarks: on the DisfluencySpeech dataset, WER dropped from 18.8% to 14.6%
Requirements: trained on Apple M4 (MPS) in fp32, 363M parameters total
Limitations: real conversational speech with disfluencies was tested only in English; other languages were evaluated on synthetic data
License: Apache-2.0
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 1.1 GB · ready files1.8 GB
  • FP8 / INT8recommendedweights 0.6 GB1.1 GB
  • 4-bitweights 0.3 GB0.9 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # Lucasgitguy-co/Fleet-Clean-0.36B
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download Lucasgitguy-co/Fleet-Clean-0.36B --local-dir ~/models/Fleet-Clean-0.36B
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Fleet-Clean-0.36B
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

  1. whistlefamily root
  2. whistle-finetuned-ko
  3. Fleet-1
  4. Fleet-2
  5. Fleet-Clean-0.36B

New posts about versions of whistle (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 5All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 2All on Hugging Face

About the model

Lucasgitguy-co/Fleet-Clean-0.36B
Author
Lucasgitguy-co
Category
Audio
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments

Audio

Fleet-2

Fleet-2