STT Console

A terminal UI for real-time speech recognition and translation, powered by Azure AI Speech. It shows live recognition and translation as you speak, and keeps a scrollable log of confirmed results.

*The audio in this screenshot comes from a public YouTube video.

Features

• Live speech recognition and live translation panes
• Confirmed-text pane that can toggle between recognized text and translated text
• Audio input selection
  • Windows: microphone or PC output (loopback) via WASAPI
  • Other platforms: the default microphone through the Speech SDK
• Source / target language selection
• Optional 2-pass recognition (PostRefinement)
• Two authentication methods
  • API key
  • Microsoft Entra ID sign-in (device code flow)
• Confirmed results are appended to a log file

Requirements

• An Azure AI Speech resource (endpoint/region and either an API key or Entra ID access)

The release builds below include the .NET runtime, so nothing else needs to be installed.

Download

Pushing a version tag (v*, for example v1.0.0) publishes a GitHub release with self-contained single-file builds that do not need the .NET runtime:

Platform	Archive
Windows (x64)	SttConsole-<version>-win-x64.zip
Windows (Arm64)	SttConsole-<version>-win-arm64.zip
Linux (x64)	SttConsole-<version>-linux-x64.tar.gz
Linux (Arm64)	SttConsole-<version>-linux-arm64.tar.gz
macOS (Apple Silicon)	SttConsole-<version>-osx-arm64.tar.gz

The macOS and Linux builds are not signed. On macOS, remove the quarantine attribute before the first run:

xattr -d com.apple.quarantine ./SttConsole

Build and run

Building from source requires the .NET 10 SDK.

dotnet run --project src\SttConsole.csproj

Usage

1. Open the settings dialogs from the top pane and configure:
   • Audio input: the device to capture from
   • Speech service: endpoint, region, and the authentication method
   • Languages: recognition source language and translation target language
   • 2Pass: toggle post-refinement on or off
2. Press S (or the Start button) to start recognition. Press it again to stop.
3. Press Q or Ctrl+C to quit.

Key bindings

Key	Action
S	Start / stop recognition
T	Switch the confirmed-text pane between recognized and translated text
Up / Down	Scroll the confirmed-text pane by one line
PageUp / PageDown	Scroll the confirmed-text pane by one page
Q / Ctrl+C	Quit

Esc is intentionally ignored so the app does not exit by accident.

Speech service settings

Open Speech Service Settings to configure the connection. Values are saved to settings.json when you press OK.

Common fields

Field	Description	Example
Endpoint URL (optional)	Endpoint of your Speech resource. Leave it empty to connect through the region. Required for Entra ID sign-in when no Speech Resource ID is set (see below).	https://my-speech.cognitiveservices.azure.com/
Region	Azure region of the Speech resource. Required unless a full endpoint URL is used with an API key, and always required when a Speech Resource ID is set.	japaneast
2-Pass Transcription	Turns on post-refinement of recognition results.
Authentication	API Key or Managed Identity (see below).

The region is the short name shown in the Azure portal (for example japaneast, eastus, westeurope), not the display name (Japan East).

Authentication: API Key

Field	Description	Example
API Key	One of the keys on the Keys and Endpoint page of the Speech resource. The input is masked.

With an API key, the app connects to the endpoint URL if one is set, and to the region otherwise. The simplest setup is therefore a region plus an API key.

Authentication: Managed Identity (Microsoft Entra ID)

Despite its name in the dialog, this option signs in as a user with your own Entra ID account. It does not use an Azure managed identity. Press Sign In in the dialog to authenticate. The status shows Signed in once it succeeds, and Sign Out clears the session.

Field	Description	Example
Tenant ID	Directory (tenant) ID of the Entra ID tenant.	00000000-0000-0000-0000-000000000000
Client ID	Application (client) ID of an app registration in that tenant.	00000000-0000-0000-0000-000000000000
Redirect URI	Redirect URI for the browser sign-in. It must match one registered on the app registration. Defaults to http://localhost:60916.	http://localhost:60916
Speech Resource ID (Entra ID)	ARM resource ID of the Speech resource. See below.	/subscriptions/<subscription-id>/resourceGroups/<resource-group>/providers/Microsoft.CognitiveServices/accounts/<speech-resource-name>
Device Code Flow	Signs in by entering a code on another device or browser instead of opening a browser from the app. Useful over SSH or on a machine without a browser.

How the app connects with Entra ID depends on the Speech Resource ID:

• Speech Resource ID set: the app uses the Region together with the resource ID and sends an aad#<resourceId>#<token> authorization token to the Speech SDK. A custom domain is not required, and the Endpoint URL is ignored.
• Speech Resource ID empty: the app connects to the Endpoint URL with the Entra ID credential. The Speech resource must have a custom domain, and the Endpoint URL is required.

Setup on the Azure side:

1. Register an application in Microsoft Entra ID and add http://localhost:60916 as a redirect URI on the Mobile and desktop applications platform.
2. If you use Device Code Flow, enable Allow public client flows on the app registration.
3. Grant the signed-in user a role that can call the Speech resource, such as Cognitive Services Speech User, on that resource.

You can find the resource ID on the Speech resource's Properties page in the Azure portal.

Files written by the app

Content	Windows	Linux	macOS
Settings (settings.json)	%APPDATA%\SttConsole\	~/.config/SttConsole/	~/Library/Application Support/SttConsole/
Entra ID authentication record (auth-record.json)	%APPDATA%\SttConsole\	~/.config/SttConsole/	~/Library/Application Support/SttConsole/
Confirmed recognition results (recognition_*.log)	%TEMP%\sttconsole\logs\	/tmp/sttconsole/logs/	$TMPDIR/sttconsole/logs/

The log file is only created when there is at least one confirmed result. Speech is sent to Azure while recognition is running, so stop recognition when you are not using it to avoid unnecessary charges.

Platform notes

• The project targets net10.0 and runs on Windows, Linux, and macOS.
• Capturing PC output (loopback) and choosing an audio device are Windows-only.
• On Linux, the Entra ID token cache may require libsecret.
• Recognition has been verified on Windows against a real Azure environment. On Linux the app starts and exits correctly, but recognition itself has not been verified.

Project layout

src/
  Models/      Settings, language options, recognition result types
  Services/
    Settings/  Settings persistence
    Auth/      Entra ID sign-in
    Speech/    Azure Speech recognition and translation
    Audio/     Audio input abstraction (WASAPI / default microphone)
  UI/          Terminal.Gui main window and dialogs
  Program.cs   Entry point and dependency injection setup
docs/images/   Screenshots used in this README

License

MIT