Repository navigation
Voice and Microphone
Özer Özdaş edited this page Oct 3, 2026
·
1 revision
Sprigo listens to you (speech recognition) and talks back (text to speech). Both are set in Settings, and Try it checks them with your own voice.
| Engine | Where it runs | Notes |
|---|---|---|
| Whisper (default) | In the app | whisper.cpp with a bundled model. Nothing to install. Uses the GPU on Apple silicon. |
| Deepgram | Cloud | Needs a Deepgram API key. Audio is sent to Deepgram. |
Sprigo never records or keeps your audio. Voice analysis measures timing (how fast you answer, pauses, answer length) from timestamps and transcripts only, and can be turned off.
| Voice | Where it runs | Languages |
|---|---|---|
| System voice | Your OS | Whatever voices your system has installed |
| Piper | In the app | Every course language. Fast. Downloaded on first use. |
| Kokoro | In the app | English only. The most natural local voice. Downloaded on first use. |
| ElevenLabs | Cloud | Needs an API key. Most natural overall. |
| Local speech server | Your own server | Any OpenAI-compatible /v1/audio/speech server, such as Kokoro-FastAPI. |
If the selected voice can't speak (for example, the language isn't supported or the download failed), Sprigo falls back to the system voice and tells you once per session.
- macOS: System Settings → Privacy & Security → Microphone → turn on Sprigo.
- Windows: Settings → Privacy & security → Microphone → allow desktop apps.
- Linux: check that your input device is selected in your sound settings (PulseAudio/PipeWire).
In a live lesson you can talk over the tutor to interrupt. If the tutor stops on its own, the microphone is probably hearing your speakers: use headphones, or turn off Settings → Cut in on the tutor.
- Turn silent mode off on iPhone, or you won't hear the tutor.
- Kokoro isn't available on the phone (it runs out of memory in mobile Safari). Use Piper or the system voice.
Using Sprigo
Contributing