# Audio generation with dialog.to_audio(): `pip install sdialog[audio]`
soundfile
pyroomacoustics
huggingface_hub
# librosa is used by `datasets` to decode the voice database audio
datasets<=2.21.0
librosa
# Dia2, the default TTS engine, requires Python >= 3.10 (on Python 3.9, other engines can be used via PerTurnTTS)
torch>=2.8.0; python_version >= "3.10"
transformers>=4.55.3; python_version >= "3.10"
safetensors>=0.5.3; python_version >= "3.10"
sphn>=0.2.0; python_version >= "3.10"
whisper-timestamped>=1.14.2; python_version >= "3.10"
