NiceVois is an experimental public service that trains portable RVC voice-model files from audio supplied by its users. The service is provided as-is during this beta and may be interrupted, changed, or withdrawn.
Your responsibility
You may upload and train only a voice you own or control, or a voice whose speaker has knowingly given you permission to create a reusable voice model. Where applicable law requires a written or otherwise specific release from the speaker, you must obtain and keep that release. You are responsible for the audio, the resulting model, and every use of that model. Do not use the service for deception, fraud, impersonation, harassment, unlawful content, or infringement of another person's privacy, publicity, copyright, or other rights.
Standing voice-training agreement
When you accept this agreement, you agree to these Terms and the Acceptable Use Policy, authorize NiceVois and its service providers to process audio you submit to create the model you request, and promise that every voice you submit meets the permission rule above. Your acceptance applies to future NiceVois training requests made through the same account until this agreement materially changes. Asking NiceVois to train a particular file, including through a connected agent, identifies the audio for that job and applies this authorization to it. NiceVois records the agreement version, acceptance time, connection type, and a pseudonymous account identifier. You will be asked to accept again if the agreement materially changes.
Access and pricing
NiceVois includes free training runs for new visitors and verified accounts, subject to displayed limits and anti-abuse controls. Stem separation, noise removal and reverb removal are provided free of charge to everyone, with no usage allowance and no account required; NiceVois may change this for future use, and doing so does not affect work already completed. Each account also receives a free voice-conversion download allowance. An OAuth-connected coding agent counts as a connected account. The NiceVois subscription includes voice training, voice conversion, stem separation and audio cleanup without a monthly usage cap, subject to the fair-use limit below; NiceVois may also contact an account whose use is materially inconsistent with individual use. Each conversion, separation, or cleanup job uses the source audio’s length once, when its first output is downloaded; listening does not. Every other output of that same job, and any repeat download of it, is included at no further cost. Subscriptions renew until cancelled; cancellation leaves access available through the paid period. Non-expiring prepaid training balance purchased before 29 August 2026 remains in the account that bought it, does not expire, and is still spent against the same displayed workload quotes; NiceVois no longer sells new balance. Jobs receive a fixed quote before training starts: voice training is quoted from audio duration and epochs. Lemon Squeezy is the merchant of record. NiceVois reserves the applicable allowance before cloud work, captures it after successful completion or an authorized download, and releases it if the job fails or is cancelled before completion. Payment refunds are governed by the Refund Policy. All jobs remain subject to displayed audio, epoch, capacity, retention, rate, and service-budget limits. Payment does not guarantee immediate GPU capacity or successful completion.
Fair use of the subscription
To keep the service available to every subscriber and to prevent abuse of shared GPU capacity, subscription voice training is limited to five hours of training time in any rolling 24-hour period, measured by each training’s estimated processing time when it starts. What five hours buys depends on the recording and the epochs you choose. At 500 epochs, the maximum, it is about seven models from three minutes of audio each, or one model from up to twenty-seven minutes of audio; a longer recording fits at fewer epochs, roughly 455 epochs for thirty minutes of audio. These are estimates, and the workspace shows the exact number of epochs that fits before a training starts. A training that does not fit in the time remaining will not start; it can be started with fewer epochs, or once earlier trainings fall outside the 24-hour period, and the workspace shows when. A training pass is bought separately, runs outside this limit and does not count towards it. Voice conversion, stem separation and audio cleanup are not affected. If you need more training time, contact info@nicevois.com.
Sharing and referral rewards
The NiceVois sharing and referral reward programme closed to new claims on 29 August 2026. No further reward credit is granted for public posts or for referred sign-ups, and public mention claims are no longer accepted. Reward credit already granted before that date remains in the account that received it, does not expire, and is used through the ordinary NiceVois training-balance system exactly as before; closing the programme is not a withdrawal of credit already given. Reward credit is not cash and cannot be withdrawn. Claims that were rejected, reversed, or still under review at closing receive no credit. You remain free to write about NiceVois, and nothing is owed either way when you do.
Your files
You retain responsibility for your source audio and downloaded model. The service does not claim ownership of them. Download your completed package before its displayed expiry date; beta files may be permanently deleted after that date.
Enforcement
Access may be suspended and content may be removed when necessary to protect people, the service, or the beta budget, or when these terms are violated. If you believe the service is being misused, contact the service operator.
Liability
To the maximum extent allowed by applicable law, the beta is provided without warranties and its operator is not liable for lost files, unsuccessful training, misuse of a model, or indirect damages. Stop using the service if you do not accept these terms.