Lolaby is an AI-powered tool designed to help children fall asleep by generating personalized lullabies. Parents and caregivers can input a child's name, age, interests, fears, and mood, and the app creates a unique song tailored to those details. The tool uses local models to generate lyrics and music, ensuring no data leaves the device. This approach prioritizes privacy and efficiency, making it ideal for use in settings like kindergartens where data security is crucial. The app's design aims to create emotionally specific lullabies that resonate with children, offering a more personalized alternative to generic sleep aids. Source: huggingface
Lolaby's pipeline begins with a drawing or typed description of what the child loves, which is analyzed by MiniCPM-V 4.6, a 1.3B vision-language model. If no drawing is provided, the app defaults to a text-based input. The model converts the drawing into a short text description that feeds into the lyric generation process. The lyrics are then crafted by a fine-tuned Llama 3.2 3B model, which generates structured lullaby lyrics with tempo, chord progression, and inline chord markers. The lyrics stream in real-time, allowing teachers to read aloud before the song is complete. Source: huggingface
The app's music is synthesized in real-time using custom DSP techniques, avoiding prerecorded samples and MIDI libraries. Instruments like guitar, piano, xylophone, harp, and ocarina are generated with harmonic profiles that create a warm, non-stimulating sound. Kokoro TTS is used for the voice, providing a calm and understated delivery that aligns with the goal of reducing stimulation during bedtime. The entire process runs on CPU, without GPU or cloud dependencies, ensuring low latency and privacy. Source: huggingface
