Next-generation AI audio generation toolkit built on PyTorch, making music creation accessible to everyone
Turn lyrics or an idea into polished AI music and singing voices—free to start.
Please wait while the demo prepares.
InspireAudio provides comprehensive audio generation solutions
Comprehensive solution for music, song, and audio generation through advanced audio tokenization technology
Fine-grained control over generation with text prompts, music genres, and structural definitions
Convenient model fine-tuning and inference scripts to meet your specific needs
Supports 24kHz mono high-quality audio output, suitable for professional scenarios
One-click installation via pip:
pip install inspire-audioGet powerful pre-trained models for InspireAudio:
from inspire_audio import download_models
download_models('inspire-audio-base')Generate music with just a few lines of code:
from inspire_audio import AudioGenerator
generator = AudioGenerator('inspire-audio-base')
audio = generator.generate(
prompt="A peaceful piano melody",
duration=30 # seconds
)Save the generated audio to a file:
audio.save('output.wav')Achieve fine-grained control with more parameters:
generator = AudioGenerator('inspire-audio-base')
audio = generator.generate(
prompt="Energetic rock music",
genre="rock",
bpm=120,
duration=60,
structure="intro-verse-chorus-verse-chorus-outro"
)InspireAudio offers various models to meet different needs
| Model Name | Description |
|---|---|
| InspireAudio-Base | Pre-trained music generation base model, 24kHz mono |
| InspireAudio-1.5B | Advanced music generation model with 1.5B parameters, 24kHz mono |
| InspireSong-Base | Pre-trained song generation base model, 24kHz mono |
Audio has a job to do in the project where it will be heard. Before writing a prompt, decide whether you need an instrumental background, a musical idea to develop, or material for a supported continuation experiment. Describe the setting and the role of the sound in a sentence. That brief becomes your reference when comparing results.
Open inspireaudio.ai to explore the toolkit overview. Use the official project resources to confirm the installation, model, and inference path for the task you want to try. Keep a small first example with a clear input and an obvious listening goal. Save the text, selected model, and relevant settings alongside the output so that a promising result remains understandable later.
Start with the main instrument or texture, then add the mood and energy you need. Avoid combining several contradictory directions in the same brief. If you want a restrained background, say what it should leave room for, such as narration or dialogue. If you want a more prominent musical passage, explain which moment it should support in the surrounding project.
Treat a prompt as a starting point for exploration. The same description can lead to different musical interpretations. Listen for whether a candidate meets the purpose of the brief rather than whether it repeats every word literally. Keep notes about useful qualities so that the next revision has a specific direction.
Imagine an instructional video with a spoken explanation. Try a brief for a calm accompaniment, then place the generated sample under a rough narration track. Listen for melodic phrases or percussion that interrupt the words. A less elaborate arrangement may work better in this context. Evaluate the combination at ordinary listening volume instead of choosing the most dramatic isolated sample.
For a short sequence that moves from stillness to activity, describe the intended emotional change and the sound palette. Review whether the candidate offers a usable opening and transition. Compare it against the timing of your edit, and mark the part that fits best. Generation is one step in the workflow; selecting, arranging, and balancing the material remain creative decisions.
If the selected model and workflow support continuation, begin with source material you are allowed to use. Listen closely around the joining point. Check whether the pulse and texture remain coherent, and whether a change feels intentional rather than accidental. Keep the original source unchanged so that you can compare attempts and return to the starting point when necessary.
Review the full output before deciding that an experiment succeeded. Listen for abrupt endings, changes in level, unwanted repetition, or details that distract from the intended purpose. A strong first few seconds may not represent the entire passage. Mark useful sections and write down what would need further editing before the audio could fit your project.
For comparisons, use the same listening setup and a similar playback level. Change one part of the prompt or configuration at a time, and save each candidate under a recognizable name. A small set of clear examples is more useful than a large folder whose inputs and settings have been lost.
A local model test, a hosted demo, and a finished production pipeline have different conditions. Record the environment when you measure a local run, and verify the requirements of the release you use. Avoid planning around a performance figure from a different device or input. The practical question is whether your chosen setup supports the kind of iteration your project needs.
If the demo is unavailable, consult the linked project resources or inspect the separately linked music service. Confirm the current model terms and the permissions for any source audio or lyrics before publication. Keep those decisions separate from your listening review: an attractive result still needs to fit the intended workflow and permitted use.
InspireAudio is an AIGC toolkit built on PyTorch that enables music, song, and audio generation through advanced audio tokenization technology.
Currently supports text-to-music generation and music continuation, with 24kHz mono high-quality audio output.
You can precisely control the output through text prompts, specifying music genres, defining music structures, and more.
We offer InspireAudio-Base, InspireAudio-1.5B music generation models, and InspireSong-Base song generation model.
Yes, InspireAudio provides convenient model fine-tuning and inference scripts, allowing you to adjust models according to specific needs.
You can contribute by starring the GitHub repository, reporting issues, submitting PRs, or joining community discussions.
Join the growing InspireAudio community! Support our development by starring our GitHub repository! 🌟