Voice
Keyboard shortcuts shown below are Windows defaults. On macOS, check the shortcut shown in TalkBar or its settings; configurable shortcuts can be changed there.
Settings → Voice controls how TalkBar records audio, turns it into a transcript, runs dictation, and labels speakers. It has four inner tabs: Recording, Transcription, Dictation and Speakers. Speech models are downloaded and removed under Settings → Private Local AI; the Manage Model link next to a model opens that page.
Recording
For how recording works end to end, see Recording and transcription.
Recording
| Setting | What it does | Default |
|---|---|---|
| Audio source | What TalkBar records: Microphone, System audio (what plays through your speakers or headphones), or Microphone + system audio. Options only appear when the device is available. The refresh button next to it checks for devices again. | Microphone |
| Microphone | The default microphone for sessions, dictation (when dictation follows it) and the audio test. Disabled when Audio source is System audio. | System default |
| Test audio input | Start test records a short sample from the current source so you can see the level and play it back. Stop test ends it. | — |
| Mute microphone shortcut | A global shortcut that mutes all microphone input to TalkBar. Dictation and recordings hear silence until you press it again. Other apps and the system microphone are not affected. The switch turns the shortcut on or off. | Ctrl+Shift+M, on |
| Show all audio devices | Lists every microphone and system-audio device TalkBar can see. Useful when a device is missing. | Off |
Processing
| Setting | What it does | Default |
|---|---|---|
| Enhanced audio processing | Applies noise suppression, gain control and echo cancellation to microphone audio. | On |
| Auto level balance | In Microphone + system audio mode, raises quiet sources and lowers loud ones automatically. | On |
Session end
| Setting | What it does | Default |
|---|---|---|
| Notify before a session auto-ends | Shows a desktop notification when an unattended session is about to stop on its own. | On |
| Auto-pause | Pauses a live session after sustained silence and resumes when someone speaks. Options: Off, 10 seconds, 15 seconds, 30 seconds. | 10 seconds |
| Auto-stop | Completes an auto-paused session after TalkBar has been unattended. Options: Off, 2 minutes, 3 minutes, 5 minutes, 10 minutes. Available only while Auto-pause is on. | 3 minutes |
Saved recording
| Setting | What it does | Default |
|---|---|---|
| Quality | Bitrate of the audio TalkBar keeps on this computer. Higher means larger files. Options: Compact · 24 kbps, Standard · 32 kbps (default), High · 48 kbps, Very high · 64 kbps, Maximum · 96 kbps. | Standard · 32 kbps (default) |
| Delete audio when a session ends | Removes the recording after you stop. The transcript, timeline and summaries stay, but playback and Enhance transcript are no longer available for that session. | Off |
Transcription
Speech-to-text runs on this computer. A model downloads when you select it; a status line under each model shows whether it is ready and offers a download button if it is not.
Transcription
| Setting | What it does | Default |
|---|---|---|
| Live transcription model | The model that writes the live transcript while you record. A fast model keeps words appearing as you speak; TalkBar warns you if you pick a slow one. | Parakeet V3 |
| After-recording enhancement model | The model used by Enhance transcript after a recording ends. Slower, more accurate models are fine here. | Whisper Turbo |
| Transcription language | The spoken language for live and enhanced transcripts. Leave it on auto-detect, or pick a language if words come out wrong. The list only offers languages your selected models support, and it is locked when both models detect the language on their own. | Auto-detect |
| Speaker detection model | The model that separates speakers in new sessions, imported audio and Enhance. Options: Automatic (Sortformer up to 4 speakers, Nemotron 3 for more), Sortformer (4 or 8 speakers), Nemotron 3 (up to 8 speakers). For more than 4 speakers, Automatic uses Nemotron 3 when it is installed, otherwise the 8-speaker Sortformer model. | Automatic (Sortformer up to 4 speakers, Nemotron 3 for more) |
The model lists include Parakeet V3, Orukeet INT8, Whisper Small, Whisper Turbo, GigaAM v3, Canary 180M Flash and Canary 1B v2. Each entry shows a short hint and its download size. See Local models for more about them.
Timelines
| Setting | What it does | Default |
|---|---|---|
| Timeline frequency | How often the live timeline (short bullets about what was said) refreshes during a session. Options: 10, 20 or 30 seconds, or 1 min. | 10 seconds |
Summaries
| Setting | What it does | Default |
|---|---|---|
| Default summary length | The summary length new sessions use: Short, Standard or Detailed. You can still change it on a single session. | Standard |
| Summary effort | How much work the AI puts into splitting and reading the transcript: Fast, Balanced or Thorough. Thorough uses more AI requests. This does not change model token limits. | Balanced |
| Run summary parts in parallel | Lets cloud and CLI models work on up to 3 parts of a long summary at once, so it finishes faster. Built-in models always work one part at a time. | On |
Dictation
Dictation lets you hold a shortcut and speak into a note, or into a floating panel in any other app. See Dictation for how to use it.
Dictation
The switch in the section header turns dictation on or off. When it is off, the other dictation controls are disabled.
| Setting | What it does | Default |
|---|---|---|
| Dictation (switch) | Turns dictation on or off. | On |
| Dictation shortcut | Hold this shortcut to start or stop dictation; a short tap does nothing. The switch turns the shortcut on or off, and Hold for sets how long you must hold it (200–3000 ms). | Ctrl+Shift, on, hold for 500 ms |
| Insert while recording | A shortcut that inserts what you have dictated so far and keeps recording. The switch turns it on or off. | Ctrl+Shift+Enter, on |
If TalkBar cannot detect the shortcut normally, a notice appears under it explaining why.
Polish with AI
The switch in the section header turns polishing on or off. When it is on, AI cleans up the transcript before it is inserted.
| Setting | What it does | Default |
|---|---|---|
| Polish with AI (switch) | Cleans up dictated text with AI before inserting it. | Off |
| Polish shortcut | Polishes and inserts in one step. Works only while Polish with AI is on. The switch turns the shortcut on or off. | Ctrl+Shift+L, on |
| Auto-polish while dictating | Cleans finished sentences in the background while you keep talking. Options: Off, or every 2, 3, 4, 5 or 8 sentences. | Off |
| Polish instructions prompt | The instructions the AI follows when polishing. Edit to change them, Save to keep your version, Revert to drop unsaved edits, Reset to default to go back to the built-in prompt. | Built-in prompt |
Overlay
| Setting | What it does | Default |
|---|---|---|
| Panel auto-expand | The dictation panel starts at 2 lines and grows up to this limit: 2, 4, 6, 8 or 10 lines. | Up to 6 lines of text |
| Remember overlay position | Reopens the panel where you last dragged it. | On |
| Default position | Where the panel opens when no position is remembered: Bottom center (default), Bottom left, Bottom right, Top center, Top left, Top right. | Bottom center (default) |
| Edge offset | How far the panel sits from the screen edges (0–200 px). | 48 px |
Speech
| Setting | What it does | Default |
|---|---|---|
| Microphone | The microphone dictation uses. | Same as voice recording |
| Dictation model | The speech model for dictation. Follows the Live transcription model unless you pick another. | Same as voice transcription |
| Dictation language | Locks dictation to one language from the next dictation. Locked to auto-detect when the dictation model only detects the language itself. | Same as voice (…), showing your Transcription language |
When dictation stops
| Setting | What it does | Default |
|---|---|---|
| Output | Where text goes when you insert or stop: Paste into the active app, Copy to clipboard only, or Paste and keep in clipboard. | Paste into the active app |
| Close after insert | Inserting also stops recording and hides the panel. | Off |
| Erase button | What Esc and the eraser remove on each press: Last word, then all, Clear all text at once, or Clear all (ask to confirm). | Last word, then all |
| Dictation history | How many recent inserts the panel keeps (open them from the dictation panel logo): Off — don’t keep history, or the last 3, 5, 10 or 15 inserts. | Keep last 5 inserts |
Speakers
Translation
The switch in the section header turns live translation on or off. When it is on, TalkBar translates each speaker turn during recording and again when the transcript is enhanced. The audio must be in the selected source language. See Live translation.
| Setting | What it does | Default |
|---|---|---|
| Translation (switch) | Turns transcript translation on or off. | Off |
| From | The spoken language to translate from. | English |
| To | The language to translate into. Only pairs TalkBar has a model for are offered. | Russian |
| Transcript panel | Translates live transcript segments. | On |
| Timeline panel | Translates live timeline bullets. | On |
Each language pair needs its own translation model. If it is not installed, a Download button with the model size appears, and translation stays unavailable until the download finishes.
Speaker labels
Speaker labels are roles you can give to speakers, such as Client or Interviewer. Labels higher in the list get more attention in timelines and summaries. See Speakers.
- Add label creates a new label.
- Drag a label to change its position.
- Click a label or its pencil button to rename it; the trash button removes it.
- Built-in labels (Client, Manager, Interviewer, Guest) show a lock and cannot be renamed or removed.
New workspaces also start with Decision Maker, Expert, Support and Observer, which you can edit or remove.
Related
Templates & Skills
Reference for Settings → Templates & Skills — session templates, summary sections, note templates, skills and knowledge skills, and what you can do with each.
Exports
Export sessions to Markdown, plain text, HTML, JSON, subtitles or ZIP, to an Obsidian vault or a Notion page, by hand or automatically.