Setting Up Voice-to-Text Workflows

9 min read · March 7, 2026

Setting Up Voice-to-Text Workflows

A voice-to-text tool is only as effective as the workflow you build around it. The technology — speech recognition, AI refinement, text output — is the foundation, but the real productivity gains come from designing workflows that fit your specific tasks, configuring your setup for reliable results, and developing habits that make voice input second nature. This guide walks through the complete process of setting up voice-to-text workflows in ummless.

Step 1: Identify Your Use Cases

Before configuring anything, identify the specific tasks where voice input will replace or supplement typing. Not every writing task benefits equally from voice input. The best candidates share these characteristics:

Natural language output. Tasks that produce prose, not code syntax or spreadsheet formulas.
First-draft quality is acceptable. Situations where you'll review and potentially edit the output, not where the text must be perfect on the first pass.
Moderate length. Text between 50 and 500 words. Very short outputs (a search query) don't justify the overhead of speaking. Very long outputs (a 5,000-word document) work better when broken into sections.
Time-sensitive. Tasks where speed matters more than polish, or where the friction of typing causes you to skip the task entirely.

Common use cases for developers and knowledge workers:

Use Case	Typical Length	Frequency
Slack/Teams messages	50-200 words	Many times daily
Commit messages	30-100 words	Several times daily
PR descriptions	100-300 words	Daily
Code comments/docs	50-200 words	Several times daily
Meeting notes	200-500 words	A few times weekly
Design documents	300-1000 words	Weekly
Email responses	100-400 words	Daily
Journal/standup notes	100-300 words	Daily

Rank these by frequency and impact. The tasks you do most often with the most friction are where you should start.

Step 2: Choose and Configure Presets

Presets are the bridge between your raw speech and polished output. Each preset contains instructions that tell the AI refinement model how to process your transcript. Choosing the right preset for each use case is the single most impactful configuration decision you'll make.

Built-in Presets

ummless ships with several built-in presets designed for common scenarios:

Clean Up — Minimal refinement. Removes filler words, fixes grammar, adds punctuation. Preserves your original wording as closely as possible.
Professional — Restructures text for business communication. Formal tone, clear paragraph structure, concise phrasing.
Casual — Light editing with a conversational tone. Good for messages, chat, and informal writing.
Technical — Optimized for developer contexts. Preserves technical terminology, formats code references properly, uses precise language.
Summary — Condenses your speech into key points. Good for meeting notes and status updates.
Bullet Points — Converts flowing speech into a structured bulleted list. Useful for action items, requirements, and notes.

Matching Presets to Use Cases

Map each of your identified use cases to a preset:

Slack messages → Casual or Clean Up
Commit messages → Technical with concise output
PR descriptions → Technical
Code documentation → Technical
Meeting notes → Summary or Bullet Points
Design documents → Professional
Email responses → Professional or Casual depending on audience
Journal entries → Clean Up

Creating Custom Presets

When built-in presets don't fit, create custom ones. A good custom preset has clear, specific instructions. Compare:

Vague (less effective):

Make this sound professional and clean it up.

Specific (more effective):

Format this as a pull request description. Start with a one-sentence summary of the change. Follow with a "Changes" section using bullet points. End with a "Testing" section describing how the change was verified. Use present tense. Keep the total length under 200 words.

The more specific your instructions, the more consistent and useful the output. When crafting presets, consider:

Output format — Paragraphs, bullet points, numbered list, specific sections?
Tone — Formal, conversational, technical, friendly?
Length — Should the output be concise or detailed? Set explicit word count targets.
Structure — Should it follow a template? Specify the template.
What to preserve — Technical terms, names, specific phrases that shouldn't be altered?
What to remove — Filler words, hedging language, repetition?

Step 3: Optimize Your Audio Setup

Consistent transcription accuracy depends on clean audio input. Small adjustments to your setup can meaningfully improve results.

Microphone Selection

Your Mac's built-in microphone is adequate for quiet environments, but a dedicated microphone improves accuracy in all conditions:

USB condenser microphones (like the Blue Yeti or Audio-Technica AT2020USB+) provide excellent audio quality for desk use.
Headset microphones position the capsule close to your mouth, reducing ambient noise pickup. Good for open offices.
Lavalier microphones clip to your clothing and provide consistent positioning. Good if you move around.

The key specification is the microphone's polar pattern. A cardioid pattern picks up sound primarily from the front, rejecting noise from the sides and rear. This is ideal for voice input in most environments.

Environment

Quiet is better. Close doors, turn off fans if possible, and avoid dictating near conversations.
Consistent background noise is manageable. The speech recognizer's noise reduction handles steady sounds (HVAC, white noise) better than intermittent sounds (notifications, keyboard clicks from a neighbor).
Room acoustics matter. Hard surfaces create reflections that can confuse the recognizer. If you're in a highly reverberant room, a close-talking microphone helps more than room treatment.

System Configuration

Verify your input device in macOS System Settings → Sound → Input. Make sure the correct microphone is selected.
Adjust the input level so your normal speaking voice peaks around 75% of the meter. Too quiet and the recognizer struggles; too loud and clipping distorts the signal.
Disable noise reduction features in third-party audio software that might conflict with the system's built-in processing.

Step 4: Develop Your Speaking Technique

How you speak matters as much as what microphone you use. Develop a deliberate speaking style for dictation:

Pace. Speak at a natural conversational rate, perhaps slightly slower than your normal speech. Rushing produces more recognition errors. Aim for about 130-150 words per minute — roughly the pace of a news broadcaster.

Articulation. Enunciate clearly without over-articulating. Swallowed consonants at the end of words ("going to" becoming "gonna") reduce accuracy. You don't need to speak robotically, but clarity helps.

Pausing. Pause briefly between sentences. This helps the recognizer identify sentence boundaries and gives you a moment to organize your next thought. A one-second pause between sentences is sufficient.

Breathing. Breathe at natural pause points, not in the middle of phrases. Mid-phrase breaths can split words and confuse the recognizer.

Corrections. If you misspeak, don't try to correct mid-stream. Finish your thought, then make a note to fix it in review. Stopping and restarting disrupts your flow and produces awkward transcripts.

Step 5: Build the Habit

The hardest part of adopting voice input isn't the technology — it's the behavior change. Most people default to typing because it's familiar, even when speaking would be faster.

The Two-Week Commitment

Commit to using voice input for at least one specific task for two weeks. Choose a frequent, low-stakes task — Slack messages or daily standup notes are ideal. The first few days will feel slower than typing. By the end of two weeks, the workflow will feel natural.

Triggers and Cues

Attach voice input to existing habits. For example: "When I finish a pull request, I speak the description instead of typing it." or "Before I push a commit, I dictate the commit message." These triggers integrate voice input into your existing flow rather than requiring you to remember a new tool.

The Quick-Capture Pattern

One of the most effective patterns is quick capture: whenever you have an idea, a task, or a note, invoke ummless with the keyboard shortcut and speak it immediately. This works for:

Capturing a bug you just noticed while working on something else
Recording a decision and the reasoning behind it
Noting a follow-up task before you forget it
Drafting a response to a message you just read

The keyboard shortcut (Cmd+Shift+Space) is designed for this — invoke, speak, dismiss. Total time: 10-15 seconds for a thought that would otherwise be lost or require switching to a notes app.

Step 6: Measure and Iterate

Track your progress to stay motivated and identify areas for improvement:

Count the saves. Keep a rough tally of how many times per day you use voice input instead of typing. Increasing frequency indicates the habit is forming.

Monitor edit distance. How much do you edit the refined output before using it? If you're consistently making the same type of correction, adjust your preset to handle it automatically.

Time comparisons. Occasionally time yourself doing the same task both ways — typing versus speaking. The first few times may not show a speed advantage. After two weeks, most people find voice input is 2-3x faster for prose-heavy tasks.

Preset refinement. Review your presets monthly. As your speaking style evolves and your needs change, update your preset instructions to match. A preset that worked well initially may need tuning after a month of use.

Putting It All Together

A complete voice-to-text workflow looks like this:

You identify a task that produces natural language text.
You invoke ummless with the keyboard shortcut.
You speak your content using a deliberate, clear speaking style.
The on-device speech recognizer transcribes your speech in real time.
The transcript is sent to the AI refinement model with your selected preset's instructions.
The refined text appears, formatted and polished according to your preset.
You review the output, make any final adjustments, and use it.

Each step in this chain can be optimized independently. Better microphone improves step 4. Better presets improve step 6. Better speaking technique improves both. The workflow compounds: small improvements at each stage produce significant overall gains in speed and output quality.

Start with one use case, get comfortable, then expand. Within a month, voice input will be a natural part of how you work.