danmohad/multiLangSpeechToImage

Multiple language speech to image running locally using OpenAI's whisper and Stability AI's stable-diffusion

Open-source multi-lingual speech-to-image project

Objective:

Given that:

• Visual art is a foundational form of human self-expression
• Speech is the foundational form of human communication
• Not everyone is literate
• Not everyone is sufficiently skilled or confident to generate visual art through traditional or digital media

And in particular:

• Not everyone is a native English speaker
• The most powerful AI text-to-image generation models are based on exclusively English-language prompts

Therefore:

• This project intends to provide a means for anyone to generate visual art directly through their speech, without presumption or prejudice with regard to their native language or level of literacy.

Pipeline:

• Speech input to notebook: IPyWebRTC
• Spoken language detection: OpenAI's whisper
• Speech-to-text (speech-to-English): OpenAI's whisper
• (English) text-to-image: Stability AI's stable-diffusion
  • Locally via 🤗 Diffusers, or through DeepAI's API

Basic requirements:

• Python v3.10.6
• An account on 🤗 (Hugging Face)
  • Must accept T&C before downloading the stable-diffusion weights
• ffmpeg
  • Can install via brew, apt, conda or other package manager

Installation:

• Create and activate a fresh python v3.10.6 venv
• git clone this repository
• Install the dependencies with pip install -r requirements.txt
• Download the stable-diffusion weights
  • git lfs install
  • git clone https://huggingface.co/runwayml/stable-diffusion-v1-5