README.md

metadata

title: Kids Rhyme Studio
emoji: 🎵
colorFrom: pink
colorTo: yellow
sdk: gradio
sdk_version: 6.28.0
python_version: 3.12.12
app_file: app.py
pinned: false
license: other
short_description: Make a child's word rhyme and hear it sung as a song
preload_from_hub:
  - ACE-Step/acestep-v15-xl-turbo-diffusers

Kids Rhyme Studio 🎵

Turn a child's words into an original sung song. Enter a first name, one to ten words, and a song theme. Pick English, Spanish, Telugu, or German, then click Make my rhyme. Edit the lines if you like, then click Sing my rhyme to hear and download a WAV file.

The song is generated with ACE-Step 1.5. It creates a vocal performance with accompaniment from the supplied lyrics; this is not spoken narration over music. Generated singing can sometimes mispronounce or change a word, especially in less common languages. Listen and regenerate if needed. No speech or instrumental-only fallback is used. All entered words appear in the written rhyme; longer lists produce a longer song.

Pick a song theme

Choose Good Habits, Kindness & Sharing, Bedtime, Counting, Colors & Shapes, Animals & Nature, or Confidence & Trying Again. Each preset has a lesson and teaching lines in all four languages.

Use the optional Theme prompt to be more specific, such as “Teach washing hands before meals and putting toys away.” Choose Custom theme to supply your own topic, such as “Feeling brave on the first day of school.” Custom themes require a prompt. For the built-in writer, write supplied words and prompts in your selected language; it preserves your text without translating it. The optional AI writer is instructed to interpret your topic in the chosen language.

The selected language is passed explicitly to ACE-Step as en, es, te, or de, alongside the lyrics and an instruction to sing in that language. This configures generation; it does not verify the language of the finished audio. If you change your language, theme, words, name, or mood after generating the rhyme, click Make my rhyme again before singing.

The built-in rhyme writer needs no token. The optional AI writer gives more varied lyrics when a parent supplies HF_TOKEN; it sends the name, words, theme, and prompt to a hosted model. The singing model runs on your Space's GPU without an Inference Providers token.

Set up a Hugging Face Space

1. Create a new Space, name it kids-rhyme-studio, and choose Gradio as the SDK.
2. In Settings → Hardware, select ZeroGPU. CPU Basic cannot generate the song. ZeroGPU provides shared GPU time subject to its usage quota.
3. Upload these six files to the root of the Space: README.md, app.py, rhyme_engine.py, themes.py, audio_engine.py, and requirements.txt. The build preloads the singing model (roughly 11 GB), so the first build may take some time.
4. In the App tab, try the example words, click Make my rhyme, and then Sing my rhyme. Wait for the GPU queue and the song to finish.
5. Optional: Under Settings → Variables and secrets, add HF_TOKEN as a secret if you want the optional AI lyric writer. Otherwise the built-in writer works as is; never paste a token into source files.

For an existing Space, upload all six files and switch its hardware to ZeroGPU before testing the song. The app shows an error when sung generation is unavailable; it does not substitute spoken audio.

Run locally

Use Python 3.12 and a CUDA GPU. Install a compatible CUDA-enabled PyTorch (version 2.8 or newer) for your system, then run:

python -m venv .venv
source .venv/bin/activate
pip install gradio==6.28.0 -r requirements.txt
python app.py

Open the address printed by Gradio. The large song model downloads at first startup. A local machine without a GPU can still write rhymes, but cannot sing them.

Notes for parents

• Read and listen to the result before sharing it with a child. The simple word guard is not a complete child-safety filter, and generated audio can differ from the displayed words.
• Use a nickname. The app has no account or rhyme database. The optional AI writer is off by default and sends entered text to its hosted provider only when enabled with a configured token.
• The model's four language codes are en, es, te, and de. Singing quality and pronunciation vary by language and input.
• ZeroGPU has daily usage limits and a queue. The Space requires Python 3.12.12 (or another version currently supported by ZeroGPU), a compatible PyTorch runtime, and sufficient model download time and disk space.
• To change the optional AI writer, set LYRICS_MODEL to another supported chat model; its default is Qwen/Qwen3-4B-Instruct-2507.

Files

File	Purpose
app.py	Simple word form, editable rhyme, and song player
rhyme_engine.py	Input checks, four-language templates, optional AI lyrics
themes.py	Suggested lessons and teaching lines in four languages
audio_engine.py	ACE-Step sung song generation on ZeroGPU
requirements.txt	Song-model Python dependencies