01SPEAK
Text that comes out as voice.
Microsoft's neural edge-tts voices speak for you; if they don't answer, gTTS steps in without a word. Every line is cached, so repeating is instant — and works offline from the second time on.
CACHED ON DISK
SOUNDBOARD · WINDOWS · OPEN SOURCE
You type.
Your mic speaks.
Type a line and it comes out as speech in the call. Play an MP3 you already have. Or pick an open window and bridge that program's audio — only that one, without dragging the rest of the machine along.
Or copy and paste:
pip install -e .The window, right here in the browser — Click the tabs — this is how the program opens.
yummi.exe
OUTPUT (BECOMES YOUR MIC)
CABLE Input (VB-Audio Virtual Cable)
MONITOR (YOU HEAR)
Headphones
CTRL+ENTER
On this page it is your browser's voice speaking, just so you can feel the flow. In the program the voice is neural and every line is cached after the first time.
LIBRARY4
edge-tts · pt-BR-FranciscaNeural
STATUS / READY48 KHZ
5
dependencies, none of them heavy
157
automated tests
0
ffmpeg, zero temp files
3.10+
Python, 3.10 and up
[ THREE TABS ]
One window. All of your audio.
The two outputs picked up top — the virtual cable that becomes your mic, and the headphones where you hear yourself — apply to all three tabs at once.
02FILES
The sounds you already have.
Point at a folder and the audio shows up in the list, ready to fire with a click or with keys 1 to 9. MP3, WAV, OGG, FLAC, OPUS and AIFF, decoded straight from memory.
NOTHING IMPORTEDMP3 · WAV · OGG · FLAC · OPUS · AIFF
03BRIDGE
The audio of a single window.
Pick Brave, Spotify, whatever is playing. Yummi captures the audio of that process and its child processes and replays it live on the output — Discord pings and game sound don't tag along.
WASAPI PROCESS LOOPBACK
[ HOW THE BRIDGE WORKS ]
Per program, not per sound card.
Capture happens before the sound reaches your speakers, and it is filtered by process. That is why you can send the video playing in your browser without sending your own voice, the game audio, or the alert that popped up halfway through.
Requires Windows 10 build 20348 or newer. Any Windows 11 will do.
- 01
brave.exe
the chosen window
- 02
Yummi
captures that process only
- 03
CABLE Input
the virtual cable
- 04
Discord
hears it as a mic
The same output still plays in your headphones, so you follow what you are sending. If you pick an audio input instead of a window — stereo mix, say — Yummi refuses the outputs that would feed back on themselves and says on screen why that one was left out.
[ KEYBOARD ]
Your hand stays put.
These work with the window focused. The global ones, for firing with Yummi minimised, are written and switched off by default — install pynput and flip one key in config.json.
- CtrlEnterspeaks the text
- CtrlSsaves the text as a clip
- CtrlOopens an audio file and plays it
- Escstops playback
- F5scans for audio devices again
- 19play the first items in the focused list
- Deleteremoves the selected clip
[ VERSION 3.0.0 · MIT ]
Ready to speak with your keyboard?
Four commands and Yummi is open. The virtual cable is optional: without it the sound stays between you and your headphones.
git clone https://github.com/ayunaangel/Yummi_text-to-speech.gitcd Yummi_text-to-speechpip install -e .yummi