Hushbird stands on excellent open work. Thank you to the researchers and engineers behind these projects.
Speech recognition model by NVIDIA, used under the Creative Commons Attribution 4.0 license. Hushbird distributes a copy converted to bfloat16 precision; the model is otherwise unmodified.
Text to speech model by hexgrad, used under the Apache License 2.0.
Language model by the Qwen team at Alibaba Cloud, used for text cleanup under the Apache License 2.0. Hushbird uses a 4-bit copy converted for Apple's MLX framework by the MLX community; the model is otherwise unmodified. It is downloaded to your Mac the first time cleanup needs it and runs entirely on your Mac.
Speech recognition by OpenAI (Whisper) and Argmax (WhisperKit), used under the MIT License.
Apple's MLX array framework and the mlx-audio project power on-device inference, used under the MIT License.
Hushbird™ is a trademark of PSYT LLC. All rights reserved.