Audio and voice
Mixing and playing sound, and speaking with a local model.
Sound
audiomixer: A real-time mixer, positional audio and procedural synthesis, written in Cone with no C.audiodevice: audiomixer's output to the speakers, over SDL3's audio streams.
Speech
voiceaudio: The microphone and speaker of a spoken conversation, with voice activity detection.speechtotext: NVIDIA Parakeet speech recognition on the GPU: 16 kHz samples in, text out.texttospeech: Kokoro 82M text to speech on the GPU: English text in, 24 kHz samples out.speechkernels: The GPU kernels for Kokoro, run through gpuwork.englishphonemes: English text to phonemes for Kokoro, with its own spelling rules where misaki would call espeak.voicechat: A spoken conversation with a local Qwen3: microphone, Parakeet, Qwen3, Kokoro, speaker.