Custom Tones & Writing Styles

Relevant source files

The following files were used as context for generating this wiki page:

• apps/desktop/src/utils/tone.utils.ts
• apps/desktop/src/components/styling/AppStylingLayout.tsx
• apps/desktop/src/components/styling/ManualStylingLayout.tsx
• apps/desktop/src/components/styling/ManualAddStyle.tsx
• apps/desktop/test/evals/post-processing-eval.test.ts
• README.md

Custom Tones and Writing Styles in mausVoice are a core feature of the AI cleanup system, allowing users to define how their transcribed voice input is refined and formatted. By utilizing Large Language Models (LLMs) like Groq, the system removes filler words, applies punctuation, and restructures text according to specific stylistic guidelines or "tones" defined by the user or provided by the system.

Users can choose from predefined system styles such as "Polished," "Email," "Chat," or "Formal," or create their own custom styles to match specific workflows. These styles can be applied globally, manually toggled via hotkeys during dictation, or automatically triggered based on the specific application currently in focus.

Sources: README.md:58-60, apps/desktop/src/utils/tone.utils.ts:13-108, apps/desktop/src/components/styling/AppStylingLayout.tsx:61-72

System Tones Architecture

The system provides several built-in tones that serve as templates for LLM post-processing. Each tone consists of a promptTemplate which contains instructions for the AI on how to handle grammar, filler words, and specific formatting like lists or emojis.

Built-in Tone Definitions

Tone ID	Name	Description	Key Capabilities
default	Polished	Natural, well-written text preserving original voice.	Cleans grammar, removes fillers, keeps original meaning.
verbatim	Verbatim	Exact transcription with no editing.	Disables all post-processing.
email	Email	Professional email formatting.	Adds greetings, body structure, and sign-offs.
chat	Chat	Casual, concise messaging style.	Removes over-punctuation, keeps personality.
formal	Formal	Polished professional register.	Removes contractions, enforces precise vocabulary.

Sources: apps/desktop/src/utils/tone.utils.ts:13-108

Tone Configuration Logic

The system distinguishes between "Template" tones (instructions for the AI) and "Style" tones (user-defined instructions). The ToneConfig type encapsulates these variations:

export type TemplateToneConfig = {
  kind: "template";
  promptTemplate: string;
  systemPromptTemplate?: string;
};

export type StyleToneConfig = {
  kind: "style";
  stylePrompt: string;
};

export type ToneConfig = TemplateToneConfig | StyleToneConfig;

Sources: apps/desktop/src/utils/tone.utils.ts:223-234

Application-Specific Styling

Users can register specific desktop applications (e.g., Slack, Chrome) and assign a unique writing style to each. When an application is focused, mausVoice automatically switches to the assigned tone for all transcriptions within that context.

The AppStylingLayout handles the mapping between application target IDs and their assigned tones.
Sources: apps/desktop/src/components/styling/AppStylingLayout.tsx:23-55, apps/desktop/src/utils/tone.utils.ts:291-301

Manual Selection and Hotkey Cycling

In "Manual" styling mode, users can select one or more active styles to cycle through using global hotkeys while dictating. This allows for rapid switching between tones without leaving the target application.

Cycle Interaction Flow

AI Post-Processing Engine

The cleanup process involves building a specific prompt for the LLM based on the user's transcription, language, and selected tone. The system uses a JSON schema to ensure the AI returns a structured result.

Prompt Construction

The prompt generation logic combines the user's name, the dictation language, and the specific instructions from the tone.promptTemplate.

1. System Prompt: Defines the persona and core rules (e.g., "Clean up transcript without changing meaning").
2. User Prompt: Contains the raw transcript and specific formatting commands (e.g., "Format as Email").
3. JSON Schema Enforcement: Ensures the response is a valid object containing the result field.

Sources: apps/desktop/test/evals/post-processing-eval.test.ts:38-60, apps/desktop/src/utils/tone.utils.ts:20-50

Tone Management & UI

The ManualAddStyle component manages the user interface for selecting active tones. It allows users to toggle which styles are available for hotkey cycling and provides access to the ToneEditorDialog for creating custom prompt templates.

• Selection Constraint: At least one style must be selected at all times.
• Global/System Styles: Built-in or organization-managed styles (marked with isGlobal or isSystem) cannot be edited by the user but can be toggled as active.
• Custom Styles: Users can define new templates using the Add action, which invokes openToneEditorDialog.

Sources: apps/desktop/src/components/styling/ManualAddStyle.tsx:32-118

Summary

Custom Tones & Writing Styles transform mausVoice from a simple dictation tool into a sophisticated writing assistant. By decoupling the transcription (what was said) from the styling (how it is written), the system enables professional-grade text generation across any desktop application. The architecture supports automated context-aware switching, manual hotkey cycling, and deep customization through LLM prompt engineering.

Sources: README.md:58-60, apps/desktop/src/utils/tone.utils.ts:13-108