Practical overview
When it fits
Use it when you want a self-hosted, script-based workflow for training a lyric text generator on a selected corpus and sampling its output.
Who it is for
- Developers exploring character-level language modeling and LSTM text generation.
- Writers or researchers who want to experiment with lyric samples trained on a chosen corpus.
Documented uses
- Collect lyrics associated with selected artists into a local training file.
- Prepare lyric text as a character vocabulary for model training.
- Train an LSTM on the prepared lyrics and generate a sample from a seed phrase.
Requirements & limits
The repository is public and lists an MIT license. The README describes local script usage and does not document an account requirement.
- The workflow requires several separate stages: gathering, preprocessing, training and sampling.
- The gathering script depends on manually supplied artist IDs for SongMeanings.
- The README identifies TensorFlow 0.10.0rc0, so users should account for the age of the documented dependency when assessing setup.
- The included output example is not consistently coherent, making this better suited to experimentation than finished lyric writing.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Train and sample from a lyric corpus
Use the documented scripts to gather lyrics, prepare character data, train an LSTM model and save a generated sample.
Before you start
- Python repository and its scripts
- Artist IDs for the lyrics source, found manually
- A lyric corpus gathered or otherwise available as a local text file
- TensorFlow; the README names version 0.10.0rc0
Steps
Run gather.py with an output file and comma-separated artist IDs to collect and concatenate lyrics locally.
Run preprocess.py on the resulting text file to create its character vocabulary file.
Run train.py with the text file, vocabulary file and a model name.
Run sample.py with the model name, vocabulary file, output filename and a seed phrase.
Expected outcome
A generated lyric sample is saved to the specified output file.
Guide limitations
- The README states that SongMeanings has no API for this gathering workflow, so artist IDs must be found manually.
- The README's example output is understandable but may not make sense.
- The documented TensorFlow version is 0.10.0rc0; no compatibility guarantee is stated.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 17 October 2026
- Original website title
- tonybeltramelli/Deep-Lyrics · Official homepage title · retained in its source language