README.md

---
license: other
license_name: apache-2.0-and-gemma
license_link: LICENSE.md
base_model:
  - google/gemma-4-e4b-it-qat-q4_0-unquantized
  - google/gemma-4-E2B-it-qat-q4_0-unquantized
  - google/gemma-3-1b-it-qat-q4_0-unquantized
tags:
  - dictation
  - speech-to-text
  - text-normalization
  - grammatical-error-correction
  - gguf
  - lora
language:
  - en
---

# Utter cleanup models

The cleanup models of [Utter](https://github.com/ductai199x/utter), a local voice dictation app.
They take what the speech recognizer (NVIDIA Parakeet) heard and return the text the speaker
meant to type:
- **Punctuation and capitals** nobody said.
- **Spoken symbols and numbers** written as people type them ("four x faster" -> "4× faster",
  "about two forty" -> "~240").
- **Fillers and false starts** removed.
- **Corrections applied:** "at 2, actually 3"; "scratch that" (the last sentence); "let's start
  over" (everything).
- **Far-back edits:** "actually, change the eggs to three boxes".
- **Spoken lists as bullets,** numbered for steps.
- **Grammar fixed,** in the speaker's own words.
- **The person's own words spelled right** when they are in their dictionary.
- **Dictated text is typed, never answered:** a question or an instruction to an AI comes out
  as text.

Utter offers Quality and Light on a computer with a graphics card, and Light and Extra light on
one without; the person chooses in the app.

## Files

| File | Model | What |
|---|---|---|
| `utter-quality-v2-q4_0.gguf` | **Quality**, v2 | Gemma 4 E4B fine-tuned, Q4_0, for llama.cpp (b11535 or later); about 3.3 GB of GPU memory when the per-layer embeddings stay in system RAM |
| `adapters/utter-quality-v2/` | | Its LoRA adapter (rank 16) on the base checkpoint, for merging or further training |
| `utter-light-v2-q4_0.gguf` | **Light**, v2 | Gemma 4 E2B fine-tuned, Q4_0; about 1.8 GB of GPU memory, for machines with little of it (Utter then runs speech recognition on the CPU) |
| `adapters/utter-light-v2/` | | Its LoRA adapter (rank 16) |
| `utter-mini-v1-q4_0.gguf` | **Extra light**, v1 | Gemma 3 1B fully fine-tuned, Q4_0 (720 MB): for computers without a graphics card, and Utter's phone app. Gemma 3 terms (see License) |

v1 (`utter-quality-v1-q4_0.gguf`, `utter-light-v1-q4_0.gguf` and their adapters) stays here:
it was trained before the data was published, on pairs that overlap today's test split.

## How it must be prompted

The rules are in the weights, so the system prompt is short, and it must be exactly this:

```
Turn this dictation into the text the speaker meant to type: punctuated, formatted and corrected the way a careful writer would type it, in the speaker's own words.
```

The user message is the transcript after `Dictated: `, optionally preceded by context lines:

```
App: email
Before: "text before the cursor"
Lists: off
Dictionary: Utter, Nguyen
Dictated: <what the recognizer heard>
```

`Dictionary:` carries only the person's terms that sound like something in this dictation, not
the whole list. The reply is the cleaned text only. Thinking must be off
(`chat_template_kwargs: {"enable_thinking": false}`).

## Training

- **Method:** LoRA (rank 16, alpha 32) on Google's unquantized QAT checkpoint of Gemma 4 (E4B
  for Quality, E2B for Light), then merged and quantized to Q4_0, the format the QAT was
  trained for.
  - **Both tiers, one recipe:** one pass over the train split of
    [ductai199x/dictation-to-text](https://huggingface.co/datasets/ductai199x/dictation-to-text)
    (79,246 pairs), lr 2e-4, 1 epoch. A share of the pairs carry `Dictionary:` lines: some hold
    the right spelling of a misheard word, some near-miss words that must not be forced in.
- **Extra light:** every weight of Google's unquantized QAT checkpoint of Gemma 3 1B trained
  (not LoRA: a rank-16 adapter left the small model its habit of running on in lists), on the
  same train split, lr 2e-5, 2 epochs, then quantized to Q4_0.
- **Data:** pairs of (recognizer output, intended text):
  - **Written side:** real passages from Common Pile (Stack Exchange, GitHub, arXiv abstracts,
    Ubuntu IRC) and Wikipedia, plus lists and LLM-written everyday messages.
  - **Spoken side:** written by Gemma 4 26B-A4B in eight kinds (plain, spoken punctuation,
    disfluent, self-correction, retraction command, sentence restatement, grammar mistakes,
    far-back edit), checked by code, spoken by Kokoro TTS and recognized by Parakeet.
- **Answer keys:** fixed for typos and grammar, with each fix verified as grammar by ERRANT.
- **Held out:** the dataset splits by passage; no test or evaluation passage was trained on.

## Evaluation

**Ready to send:** the share of dictations a careful writer would send exactly as typed. Gemma 4
26B-A4B judges each one against the intended text, with a rubric calibrated
on human review.

**Test sets:**
- **Realistic:** 950 held-out dictations, everyday messages, emails, notes and AI prompts plus
  short real forum and chat posts. Each is spoken one way: plainly, with fillers, a
  self-correction, a command, a later edit, spoken punctuation or broken grammar.
- **Directed:** 292 dictations of questions, requests to AI assistants, instructions to coding
  agents, and text telling the dictation app itself to do something.

| Ready to send | Quality v2 | Light v2 |
|---|---|---|
| Realistic, Kokoro voices | 88.1% | 85.4% |
| Realistic, with a 300-term dictionary | 87.8% | 85.7% |
| Realistic, unseen Chatterbox voices | 80.1% | 75.5% |
| Directed, with a dictionary | 87.7% | 83.6% |
| Cleanup time, median (RTX 3090 Ti) | about 250 ms | about 180 ms |

**Extra light**, judged the same way on a larger held-out set (25,469 dictations), beside
Light on the same set:

| Ready to send | Light v2 | Extra light v1 |
|---|---|---|
| Kokoro voices, real passages | 85.7% | 79.7% |
| Unseen Chatterbox voices | 74.6% | 67.0% |
| Typed text test | 64.3% | 51.9% |
| Cleanup time, median (Pixel 10 Pro XL CPU) | 2.69 s | 1.49 s |

Utter retries a reply that runs on (one more pass that forbids a leading list marker); with
that retry, 1 of the 25,469 still ran on (Light: none).

**Known gaps:**
- A leading filler with a comma ("Uh, so...") can survive.
- Pronoun case ("me and him went") is not always fixed.
- Speech in the evaluation is synthetic (two TTS engines), not recorded dictation.

## License

- **Quality and Light** (Gemma 4): Apache 2.0, as their base models.
- **Extra light** (`utter-mini-*`, Gemma 3): Gemma is provided under and subject to the Gemma
  Terms of Use found at [ai.google.dev/gemma/terms](https://ai.google.dev/gemma/terms), including
  the [Gemma Prohibited Use Policy](https://ai.google.dev/gemma/prohibited_use_policy). This
  model is a modification of Gemma 3 1B; anyone who uses or redistributes it must follow
  those terms and pass them on.

The training text came from openly licensed sources (public domain, CC0, CC BY-SA, MIT,
Apache-2.0).