Scribosermo STT Setup

Scribosermo is a LGPL licensed, open-source speech recognition engine to "Train fast Speech-to-Text networks in different languages".

Evaluation tests for German language suggest that it's currently one of the fastest and most accurate open-source STT systems.

This repository trys to offer build scripts to run and test Scribosermo on different platforms focussing on Raspberry Pi SBC. Ultimately the goal is to build a module for SEPIA STT-Server.

Test Scribosermo

The easiest way to get started is to build and use the Docker container:

• Use the scripts inside the build folder. Tested on aarch64 and amd64 platforms.
• Download a model. Check tests folder for more info and licenses.
• Put the model inside a folder and share this folder with your Docker container, e.g. use a run flag similar to: -v my/model/folder:/home/admin/scribosermo-stt-setup/tests/model
• Run the container. It will automatically call the Python test script testing_tflite.py.
• NOTE: The Python test script is currently configured to use German. You may need to modify it if you change the model or language.

Build wheels on Debian 10 (the long way)

If you can't find matching Python wheel files for your build this might help to fill the missing parts:

• Install required packages: apt-get update && apt-get install -y --no-install-recommends sudo git wget curl nano unzip zip procps build-essential cmake python3-pip python3-dev python3-setuptools python3-wheel python3-venv libsndfile1
• Install Rust compiler (might be required): curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh and refresh terminal source $HOME/.cargo/env
• Create and activate Python virtual env: mkdir -p install && cd install && python3 -m venv env && source env/bin/activate
• Make sure pip is updated (tested v21.3.1): pip3 install --upgrade pip
• Install part 1: pip3 install wheel setuptools setuptools_rust transformers tqdm librosa datasets jiwer
• Install part 2: pip3 install --extra-index-url https://google-coral.github.io/py-repo/ tflite_runtime
• Install part 3: pip3 install ds-ctcdecoder==0.10.0a3;
• Create wheels as needed: pip3 wheel [package]

Credits

• DanBmh - Development and maintaining of Scribosermo
• Domcross - German STT evaluation, scripts and packages
• SEPIA Framework - Open assistant and STT server stuff