Skip to content
EchoFlow

1. Getting started

This guide covers EchoFlow v1.1.9 Beta for Windows.

  1. Download EchoFlow and run the installer. It asks you to accept the EchoFlow Terms. The Beta installer isn't Authenticode-signed yet, so Windows may show an unknown-publisher (SmartScreen) notice.
  2. Open EchoFlow. The first time, a five-step setup starts.
  3. Check your microphone: choose your input device and say a few words until EchoFlow shows “Signal detected”.
  4. Set up EchoFlow: EchoFlow checks your PC and recommends a setup, either Balanced Precision (Whisper Small with Phi-3 Mini) or Instant Velocity (Whisper Base with Gemma 2 2B). Choose one and select Download & Set Up. The voice and AI models download once.
  5. Learn your shortcut: hold Win + Alt, say something, and release. Your words appear on screen.
  6. Select Launch EchoFlow. You're ready to dictate.

You don't need to install anything else: EchoFlow includes the runtime components it needs, so there's no separate Microsoft Visual C++ or Vulkan install. You'll need a microphone, and an internet connection for the first model downloads.

If your setup includes Gemma 2, EchoFlow asks you to accept its licence before downloading it. You can change models at any time in Settings.

2. Your first dictation

  1. Click into any text field where you want to write, such as an email, a chat, a document or a code editor.
  2. Hold Win + Alt. The Aurora Pill near the bottom of your screen shows Recording and a timer.
  3. Speak naturally while you hold the shortcut.
  4. Release the shortcut when you're done. The Aurora Pill shows Transcribing, then Cleaning if AI refinement is on, then Inserting.
  5. EchoFlow types the text where your cursor is.

Recognition works best with your microphone reasonably close, little background noise and clear speech. It isn't perfect, so check important text before you send it.

If the recording is very short or silent, EchoFlow tells you and doesn't insert anything.

Win + Alt is the default shortcut. You can change it in Settings › General › Global Dictation Shortcut.

3. Direct Speech and AI refinement

EchoFlow can turn your speech into text in two ways:

  • Direct Speech inserts the transcription without AI rewriting, so it's quicker.
  • AI refinement (on by default) passes the transcript to your AI model, which tidies grammar, phrasing and tone in your chosen writing style.

Switch between them with the AI Polish button on Home, or the AI Style Refinement switch in Styles. If no AI model is downloaded, EchoFlow uses Direct Speech.

In Styles, Auto-Detect chooses a writing style from the app you're dictating into, such as email, chat, work tools or documents. Or choose Manual Presets and pick a tone for Personal Chat, Work & Teams, Email Correspondence, and Notes & Documents.

Both run on your PC. If refinement changes something you wanted to keep, use Undo AI edit in History to see your original words.

4. Speech models

EchoFlow turns speech into text with OpenAI's Whisper speech models, running on your PC. Smaller models are faster and lighter. Larger models are more accurate, but they're bigger downloads and need more of your PC's resources.

  • Tiny (75 MB): the fastest. Suits clear English on older PCs.
  • Base (145 MB): fast, for everyday clear English.
  • Small (480 MB): a balance of speed and accuracy, used in the Balanced Precision setup.
  • Medium (1.5 GB): more accurate, with stronger support for languages other than English.
  • Large V3 (3.1 GB): the most accurate, including for strong accents, Hinglish and Punjabi. EchoFlow recommends a dedicated graphics card for it.

8 GB of RAM is a reasonable baseline. For Large V3 with AI refinement running on the CPU, 16 GB is recommended.

Download, switch or delete speech models in Settings › Speech Engine, under Whisper Transcription Models.

For a side-by-side guide to every speech and AI model, see Models.

5. AI models

When AI refinement is on, an AI model running on your PC refines your transcript.

  • Auto Brain: chooses among the AI models you've downloaded, based on your PC and what you're dictating. It needs at least one AI model downloaded.
  • Gemma 2 2B (Google, 1.6 GB): the smallest and quickest.
  • Phi-3 Mini (Microsoft, 2.4 GB): a balanced choice with a modest memory footprint.
  • LLaMA 3 8B (Meta, 4.9 GB): larger, aimed at more involved writing.
  • Mistral NeMo 12B (Mistral AI, 7.1 GB): the largest, best suited to PCs with a dedicated graphics card.

Smaller models respond faster and use less memory. Larger models can handle more nuance, but they're slower, especially without a graphics card.

Each model is made by another organisation and stays under its own licence. EchoFlow shows a model's licence before it downloads, and Gemma 2 and Meta Llama 3 ask you to accept theirs first. The Terms list every model's licence.

Download, switch or delete AI models in Settings › Speech Engine, under Local Neural Brain for Adaptive Styles.

6. Languages

Choose the language you'll speak from the language menu at the top of Home, or in Settings › Speech Engine › Spoken Language.

The options are Auto-Detect, English, Hinglish (speak Hindi and English, get Hinglish in Roman script), Hindi → English (speak Hindi, get English), Hindi, Punjabi, Spanish, French, German, Chinese, Japanese, Korean and Russian.

  • Choose the language you're actually speaking. If Auto-Detect picks the wrong language, select yours from the menu.
  • Accuracy varies by language and by speech model. Larger speech models generally handle languages other than English better.
  • For Punjabi, use the Large V3 speech model. EchoFlow recommends it, and it writes Punjabi in Gurmukhi. Smaller models are less reliable for Punjabi and may write it in another script. Occasional spelling or script mistakes can still happen, and English words mixed into Punjabi are written in Gurmukhi.
  • A good microphone and a quiet room help in every language.

7. History

History lists your dictations with their times, newest first. Open it from the sidebar.

  • Search: find any past dictation.
  • Copy: copy a dictation's text.
  • Undo AI edit: on a dictation that AI refinement changed, shows your original words. Restore AI switches back.
  • Delete: removes one dictation after you confirm. Your statistics aren't affected.
  • Clear All: permanently deletes your whole History.

History stays on your PC. It keeps the text of your dictations, not audio recordings.

8. Settings and models

Open Settings from the sidebar. It has five sections:

  • General: Global Dictation Shortcut, Color Theme and Accent Color.
  • Audio & System: Microphone, Dictation Audio Sounds, Launch at Startup and GPU Acceleration.
  • Speech Engine: Spoken Language, speech models and AI models. Download, switch or delete models here.
  • Reset App: Reset Settings to Default keeps your History and models. Erase All EchoFlow Data removes your History, settings and downloaded models and starts setup again.
  • About EchoFlow: Check for Updates, the EchoFlow website, links to the Terms, Privacy Policy, third-party notices and model licences, and the Setup Wizard & Guided Tour.

Turn AI refinement on or off with the AI Polish button on Home, or in Styles.

9. Updates

  • Shortly after EchoFlow opens, it checks online for a new version.
  • If one is available, EchoFlow shows what's new, with two choices: Remind Me Later or Download & Install.
  • Nothing downloads or installs until you choose Download & Install. EchoFlow then installs the update and restarts.
  • Remind Me Later hides the prompt until EchoFlow next starts.
  • To check yourself, go to Settings › About EchoFlow › Check for Updates.
  • Updates keep your History, settings and downloaded models.

10. Troubleshooting

EchoFlow didn't understand me

  • Move your microphone closer and reduce background noise.
  • Check EchoFlow is using the right microphone in Settings › Audio & System › Microphone. The live mic level should move when you speak.
  • Choose the language you're speaking rather than Auto-Detect.
  • Try a larger speech model, such as Small or Medium.
  • If AI refinement changed your meaning, use Undo AI edit in History, or switch to Direct Speech.

Text isn't appearing

  • Click into an editable text field before you hold the shortcut.
  • Make sure EchoFlow is running: look for its icon in the Windows system tray.
  • Hold Win + Alt for the whole time you speak, then release. Very short or silent recordings aren't inserted.
  • In some apps that use Alt-key menus, Win + Alt can occasionally interfere with inserting text. Try the shortcut again.

EchoFlow doesn't hear my microphone

  • Choose your microphone in Settings › Audio & System › Microphone.
  • In Windows privacy settings, make sure “Let desktop apps access your microphone” is allowed.

A model download failed

  • During setup, select Retry Download. Cancel download stops a download in progress.
  • In Settings › Speech Engine, select Download again for that model.
  • Check your internet connection. Each model downloads once.

EchoFlow is running but I can't see it

  • Closing the EchoFlow window doesn't quit it. EchoFlow keeps running in the Windows system tray, so your shortcut still works.
  • Right-click the tray icon and choose Dashboard to open EchoFlow, or Quit to close it completely.
  • EchoFlow starts with Windows unless you turn off Launch at Startup in Settings › Audio & System.
  • Opening EchoFlow again brings up the copy that's already running, rather than starting a second one.

Transcription is slow

  • Larger speech and AI models are slower, especially without a graphics card. Try a smaller speech model, a smaller AI model, or Direct Speech.
  • EchoFlow can use your graphics card through Vulkan for faster local processing; this has been verified on NVIDIA graphics cards. Check GPU Acceleration is on in Settings › Audio & System. If your graphics card can't be used, EchoFlow runs on the CPU automatically.

Part of a long dictation is missing or repeated

  • Dictations longer than about 30 seconds can occasionally drop or repeat a sentence around the 30-second mark. This is a known issue. Until it's fixed, dictate in shorter pieces.

The clipboard or History looks wrong

  • If you had copied files (rather than text), they may no longer be on the clipboard after EchoFlow inserts text. Copy them again. This is a known issue.
  • If two dictations are saved in the same second, Undo AI edit may apply to the other one, and both may show as Copied. This is a known issue.

11. Privacy

  • Speech recognition and AI refinement run on your PC. Your dictated audio and text aren't sent to EchoFlow or a cloud AI service.
  • EchoFlow uses the internet to download models and to check for and download updates.
  • Optional analytics on this website are separate from the app and never include your dictation.

Full details are in our Privacy Policy.

12. Still need help?

Still stuck? Get in touch. Email contact@echoflow.uk or use our Contact page. EchoFlow is in Beta, so bug reports are especially useful.