AudioGPT

AudioGPT connects ChatGPT and a series of Audio Foundation Models to enable sending and receiving speech, sing, audio, and talking head during chatting.

Capabilities

Here we list the capability of AudioGPT at this time. More supported models and tasks are coming soon. For prompt examples, refer to asset.

Speech

Task	Supported Foundation Models	Status
Text-to-Speech	FastSpeech, SyntaSpeech, VITS	Yes (WIP)
Style Transfer	GenerSpeech	Yes
Speech Recognition	whisper, Conformer	Yes
Speech Enhancement	ConvTasNet	Yes (WIP)
Speech Separation	TF-GridNet	Yes (WIP)
Speech Translation	Multi-decoder	WIP
Mono-to-Binaural	NeuralWarp	Yes

Sing

Task	Supported Foundation Models	Status
Text-to-Sing	DiffSinger, VISinger	Yes (WIP)

Audio

Task	Supported Foundation Models	Status
Text-to-Audio	Make-An-Audio	Yes
Audio Inpainting	Make-An-Audio	Yes
Image-to-Audio	Make-An-Audio	Yes
Sound Detection	Audio-transformer	Yes
Target Sound Detection	TSDNet	Yes
Sound Extraction	LASSNet	Yes

Talking Head

Task	Supported Foundation Models	Status
Talking Head Synthesis	GeneFace	Yes (WIP)

Todo

• clean text to sing/speech code
• merge talking head synthesis into main
• change audio/video log output
• support huggingface space

Acknowledgement

We appreciate the open source of the following projects:

Visual ChatGPT   Hugging Face   LangChain   Stable Diffusion