metadata

license: mit
language:
  - en
pipeline_tag: text-to-audio
tags:
  - audiocraft
  - audiogen
  - styletts2
  - shift
  - audeering
  - sound
  - audio-generation
  - text-to-speech
  - mimic3

Affective TTS / SoundScapes

SHIFT TTS tool
Analysis of TTS emotionality #1
Soundscapes trees, water, .. via AudioGen
landscape2soundscape.py generates soundscape / overlays TTS / creates video from image.

Available Voices

Listen to available voices!

Flask API

Create virtualenv

Clone

git clone https://huggingface.co/dkounadis/artificial-styletts2

Install

virtualenv --python=python3 ~/.envs/.my_env
source ~/.envs/.my_env/bin/activate
cd artificial-styletts2/
pip install -r requirements.txt

Flask API

CUDA_DEVICE_ORDER=PCI_BUS_ID HF_HOME=./hf_home CUDA_VISIBLE_DEVICES=2 python api.py

Landscape 2 Soundscape

The following needs api.py to be already running on a tmux session.

# TTS & soundscape - overlay to .mp4
python landscape2soundscape.py

For SHIFT demo / Collaboration with SMB

YouTube Videos

Live Demo - Paplay

Special Flask API for playing sounds live

CUDA_DEVICE_ORDER=PCI_BUS_ID HF_HOME=/data/dkounadis/.hf7/ CUDA_VISIBLE_DEVICES=4 python live_api.py

Client - Describe any sound with words and it will be played back to you.

python live_demo.py  # will ask text input & play soundscape

Simple Demo

CUDA_DEVICE_ORDER=PCI_BUS_ID HF_HOME=/data/dkounadis/.hf7/ CUDA_VISIBLE_DEVICES=4 python demo.py

AudioBook

Convert your .docx to audio .wav. Via multiple voices, then concatenate all audiobooks.wav made with each voice to a full one concatenate audiobook has noisy speech, the individual single-voice audiobooks are clean, some issue with ffmpeg. Therefore, for now, SHIFT repo only produces single-voice audiobook. Archiving the multiple-voice audiobook.py here.

# uses Flask api.py
# needs to load ../shift/assets/INCLUSION_IN_MUSEUMS_audiobook.docx
#
# 
python audiobook.py