Skip to content

audio.cpp Model Licenses and Commercial Use

Written in September 2026. License terms change between versions, so everything below reflects the situation at that time. Always check the project's current LICENSE text before making a commercial decision.

Read this first (important disclaimer) There are a lot of models available for download here, and we recommend using them for learning and research only. We only provide the software installation service. This document makes no guarantee about the accuracy of any model's license. If you plan to use a model commercially, verify the license terms yourself through an authoritative source (the model's official repository, the original author's website or contact details) before you use it. The classifications and conclusions below come from our own reading of publicly available information at the time of writing, and may change as upstream projects update. A model marked "to be confirmed" simply means we have not verified it yet — we will keep filling these in.


The two projects covered here

audio.cpp

Repository: https://github.com/0xShug0/audio.cpp

A local audio processing tool that runs a range of audio features on your own machine, including text-to-speech, speech recognition, voice conversion, and vocal/accompaniment separation. It is written in C++ and uses quantization to optimize the models (in short: compressing them to make them lighter), so it runs considerably faster than the official apps and uses less RAM and VRAM — even a mid-spec computer runs it smoothly. That compression can very occasionally cost a little output quality, but for most users the results are perfectly acceptable. It supports a wide range of computers and GPUs, and it runs entirely locally with no internet connection required, so your privacy is protected.

audio.cpp-webui

Repository: https://github.com/kigner/audio.cpp-webui

audio.cpp-webui builds on the audio.cpp project and adds a friendly graphical interface. Instead of typing commands, you use everything through a web page — upload audio, type your text, pick your parameters, and generate the result in one click. This version pays particular attention to the Windows experience: it ships with a one-click launch script, so you just download it and double-click. It makes those fast, lightweight local AI audio capabilities easy for non-technical users too.


Understand this first: the software license ≠ the model license (the key point)

When you use audio.cpp / audio.cpp-webui, you are actually bound by two independent sets of licensing terms. Almost every commercial-use risk comes from confusing the two.

LayerWhat it isLicensing
① The software itselfThe source code and executables of audio.cpp and audio.cpp-webuiApache-2.0 — permissive open source; free to use, modify, and commercialize
② The model weightsThe TTS, ASR, voice conversion, and music generation models you download and load inside the softwareVaries model by model — anywhere from "free for commercial use" to "commercial use prohibited"; each one must be checked (see the classifications below)

The software being Apache-2.0 does not mean the models you load with it — or the content you generate — are free to commercialize. What determines your commercial risk is always layer ②: the license of the model itself.

Three things to watch out for:

  1. Quantization does not change the license. audio.cpp loads quantized, compressed model files (smaller and faster). Quantization only converts the numerical precision of the original weights — the result is a derivative work of the original model, and the original license terms continue to apply unchanged. "I'm not using the official original, I'm using a compressed version" is not a way around the license.
  2. Rights to the inference output are a separate question. Some model licenses restrict not only whether you may use the model, but also whether you may commercially use the content it generates. If your business output is the audio itself (voiceover, audiobooks, music, and so on), read the license text closely on this point.
  3. Bundled sub-components are covered too. Running a model often also loads separate tokenizers, text encoders, vocoders, and similar components. These may come from different projects under different licenses — verify them as well before commercial use.

A simple way to stay out of trouble: if you're just experimenting locally, learning, or doing research, use whatever you like. The moment any commercial purpose is involved (charging customers, internal business use, monetized content, deliverables for a client, etc.), pick only from section "1. ✅ Free for commercial use" below and keep the required attributions. For models under "⚠️ Conditional commercial use", read the full license text clause by clause before deciding. For anything under "❌ Commercial use prohibited" or "❓ To be confirmed", do not use it commercially at all until you have verified it.


Legend

MarkerMeaning
✅ Free for commercial usePermissive license; generally no fee and no extra conditions for commercial use
⚠️ Conditional commercial useUsually free for commercial use, but subject to conditions (revenue/scale caps, attribution, copyleft, registration, etc.) — read the original license first
❌ Commercial use prohibitedResearch / non-commercial use only; commercial use is not permitted
❓ To be confirmedWe have not been able to verify the license yet. This does not mean it is commercial-friendly — confirm it yourself before use

1. ✅ Free for commercial use (prefer these)

Text-to-speech (TTS)

ModelLicenseNotes
ChatterboxMITPermissive open source; free to use and commercialize
GLM-TTSMITPermissive open source; free to use and commercialize
VibeVoiceMITMicrosoft long-form multi-speaker TTS
Confucius4-TTSApache-2.0Permissive open source; free to use and commercialize
DotTTSApache-2.0Permissive open source; free to use and commercialize
MOSS-TTS-LocalApache-2.0MOSS flagship TTS
MOSS-TTS-NanoApache-2.0MOSS lightweight realtime TTS
Qwen3-TTSApache-2.0Permissive open source; free to use and commercialize
VoxCPM2Apache-2.0OpenBMB tokenizer-free TTS
PocketTTSCC-BY-4.0Commercial use allowed, attribution required per CC-BY

Speech recognition (ASR)

ModelLicenseNotes
Qwen3-ASRApache-2.0Permissive open source; free to use and commercialize
SenseVoice-SmallApache-2.0Multilingual speech recognition, speech emotion recognition, and audio event detection
VibeVoice ASRMITMicrosoft long-form speech transcription
Parakeet-TDT 0.6B v3CC-BY-4.0Commercial use allowed, attribution required per CC-BY

Speech analysis (forced alignment / speaker)

ModelLicenseNotes
Qwen3 Forced AlignerApache-2.0Audio-to-text forced alignment

Audio / music generation

ModelLicenseNotes
ACE-Step (1.5)MITMusic generation; note that ACE-Step v1-3.5B is the Apache-2.0 release — what this project bundles is 1.5 (MIT)
HeartMuLaApache-2.0Music generation (HeartMuLa-oss-3B + HeartCodec)

2. ⚠️ Conditional commercial use (you must read the specific license terms)

The models below may allow free commercial use once certain conditions are met, but every vendor's terms differ. Check the original license clause by clause before any commercial use. What follows lists only the license name and the typical restrictions — it is not a substitute for the original text.

ModelCategoryLicense nameTypical conditions (the original text governs)
IndexTTS2TTSbilibili Model Use LicenseBilibili's custom model license; typically includes restrictions on scope of use, attribution, etc.
DramaBoxTTSLTX-2 Community License (Lightricks)Community license; usually permits free commercial use subject to revenue/scale caps and similar conditions
Fun-ASR-NanoASRFunASR Model Open Source License Agreement v1.1Custom license; often free for commercial use but with scale/revenue/attribution limits and a ban on reselling the weights
Nemotron 3.5 ASRASRNVIDIA custom license (listed as "other" on HF)Non-standard open source agreement; usually permits commercial use with usage restrictions, notice-retention terms, etc.
Seed-VCVoice conversionGPL-3.0Commercial use allowed, but copyleft is viral — distributing derivative works requires releasing them under the GPL. Have legal review it before integrating into a closed-source product
Stable Audio 3Audio generationStability AI Community License (stable-audio-community)Free for research/non-commercial use; commercial use requires annual revenue under US$1M plus registration at stability.ai; at US$1M or above you must purchase an Enterprise License; may not be used to train a competing foundation model
MiniMax-H3Audio generationMiniMax H3 Community License AgreementCommunity license; generally allows limited commercial use subject to revenue/scale conditions — read the MiniMax license text for specifics

3. ❌ Commercial use prohibited (research / non-commercial only)

ModelCategoryLicenseNotes
Hviske ASRASRCC-BY-NC-4.0Non-commercial license; commercial use prohibited
OmniVoiceTTSCC-BY-NC-4.0Non-commercial license; commercial use prohibited
Sortformer DiarizationSpeech analysisCC-BY-NC-4.0Non-commercial license; commercial use prohibited
Fish Audio S2 ProTTSfish-audio-research-licenseResearch-use license; commercial use prohibited

4. ❓ License to be confirmed

We have not been able to verify the licenses of the models below, so they are grouped here as unconfirmed. This is only a temporary state — we will keep completing this document. Until a license is verified, do not assume it is commercial-friendly; if you need to use one, confirm it yourself with the upstream repository or the original author.

Text-to-speech (TTS)

Higgs Audio v3 TTS, Inflect Micro v2, Irodori-TTS, MioTTS, NeuTTS, Llama-OuteTTS 1.0, Supertonic 3, VieNeu-TTS v3 Turbo

Speech recognition (ASR)

Citrinet ASR, Higgs Audio v3 STT, Kroko Community ASR, Voxtral Mini 4B Realtime

Voice conversion

RVC, Vevo2

Audio processing tools (separation / codec / score)

BS-RoFormer, HTDemucs, Mel-Band RoFormer, MioCodec, MuScriptor


One more reminder: the audio.cpp / audio.cpp-webui software itself is Apache-2.0, but that grants you no rights over any model. Always follow the "software license ≠ model license" reasoning above and check the license of every model you actually load, along with its bundled sub-components. There are many models available for download here and we recommend using them for learning and research only; we provide the installation service alone and make no guarantee about the accuracy of any model's license. For commercial use, verify through an authoritative source such as the model's official repository or the original author's website, or contact the original developer directly.