Voice · speech ↔ text
Transcribe and speak, locally.
Whisper-grade transcription and natural text-to-speech, running on your Mac. Your microphone never leaves the machine.
What you can do
- Speech-to-text — mic or file, with timestamps
- VAD & diarization — detect speech, label speakers
- Natural TTS — read responses aloud, choose voices
- Streaming — live transcription & synthesis
Models
The best open voice models, MLX-accelerated.
- Whisper / Distil-Whisper / Parakeet (STT)
- Kokoro / Qwen3-TTS (TTS)
- Speech-to-speech where supported
- Voice selection & streaming
Powered by mlx-audio. See the voice docs.
How it works
OpenAI-compatible audio endpoints, on localhost.
POST localhost:1337/v1/audio/transcriptions # speech → text POST localhost:1337/v1/audio/speech # text → speech GET localhost:1337/api/v1/audio/voices # list voices
FAQ
Can I transcribe audio locally on a Mac?
Yes. Bedrova runs Whisper and other speech-to-text models on Apple Silicon, from your microphone or a file, with timestamps and segments. Voice activity detection and speaker diarization are available where the model supports them. All of it is on-device.
Does it do text-to-speech too?
Yes. Natural voices through Kokoro, Qwen3-TTS and others, with voice selection, custom voices and streaming synthesis that starts playing before the whole clip is done.
Does my audio leave the Mac?
No. Microphone and file audio stay local by default; nothing is uploaded.