Practical overview
When it fits
Use it when you need a research codebase for experiments on automatic transcription of sung lyrics and can prepare the required audio datasets and software environment.
Who it is for
- Researchers evaluating automatic lyric transcription methods on singing datasets.
- Developers building or adapting a research prototype for sung-lyric transcription.
- Practitioners reproducing the repository's documented dataset experiments.
Documented uses
- Adapt wav2vec 2.0 speech-recognition methods to singing audio.
- Train or evaluate lyric-transcription experiments on prepared datasets.
- Compare transfer-learning starting points and the hybrid CTC/attention approach.
Requirements & limits
The repository is public and identified as open source under the Apache-2.0 license.
- The README requires mono audio sampled at 16 kHz.
- Documented environment versions include Python 3.8.12, PyTorch 1.9.1 and CUDA 11.1.
- Dataset setup is part of the workflow; the README gives structures and metadata expectations for DSing, DALI, Hansen, Jamendo and Mauch.
- DALI preparation uses Demucs v3 with the mdx_extra model for source separation before training and evaluation.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Prepare the documented experiment environment
Set up the Python and package environment specified for ALT_SpeechBrain.
Before you start
- Anaconda
- Python 3.8.12
- PyTorch 1.9.1 with CUDA 11.1, torchvision 0.10.1 and torchaudio 0.9.1
- A local checkout of ALT_SpeechBrain
Steps
Create the Conda environment named alt with Python 3.8.12.
Install the specified PyTorch, torchvision and torchaudio packages using the CUDA 11.1 wheel index.
From the ALT_SpeechBrain directory, install requirements.txt and the project in editable mode.
Install Transformers, Datasets and scikit-learn as listed in the README.
Expected outcome
The README's specified software environment and project dependencies are installed.
Guide limitations
- This covers environment setup only; it does not establish that an experiment runs successfully.
- The repository's instructions specify a CUDA 11.1 environment.
Run a dataset experiment using its recipe
Launch a documented training or evaluation experiment after preparing the dataset/task files.
Before you start
- Installed ALT_SpeechBrain environment
- A prepared supported dataset and its dataset/task directory
- Audio at 16 kHz, mono-channel
- The task's params.yaml and any required CSV files
Steps
Prepare the dataset using the structure and metadata guidance for the chosen dataset/task.
Change into the relevant dataset/task directory.
Run the command below.
python experiment.py params.yamlCheck the output_folder specified in the YAML file for results and detailed logs; less verbose logs are printed to standard output.
Expected outcome
The experiment is configured to save its results and detailed logs in the YAML-specified output folder.
Guide limitations
- The README says CSV files may need to be created according to guidance within each dataset/task directory.
- This recipe reports the documented procedure, not a verified successful run.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 16 October 2026
- Original website title
- guxm2021/ALT_SpeechBrain · Official homepage title · retained in its source language