# NiceVois > NiceVois is the simple, one-tap way to train a portable RVC v2 voice model from permitted audio. Upload the audio, choose epochs, name the model, and press Start Training; NiceVois handles the rest and produces a named .pth, matching .index, and complete ZIP package. NiceVois also works as an RVC voice converter and vocal cleaner. Customers can use a voice trained in their NiceVois library or upload a private RVC .pth or complete model ZIP. One conversion returns the normal RVC output and a NiceVois Clean Consonants output that reduces robotic sibilance and restores natural s, sh, f, t, and breath sounds. ## Simple customer workflow 1. Upload WAV, MP3, FLAC, M4A, or OGG voice audio. 2. Choose the epochs and name the model. 3. Press Start Training once. 4. Download the finished .pth, matching .index, or complete ZIP. No local RVC installation, notebook, command line, or local GPU is required. ## Use the tool - [All NiceVois tools](https://nicevois.com/tools) - every tool in one place, with the pages about each - [Train an RVC model](https://nicevois.com/) - [Audio to PTH explained](https://nicevois.com/audio-to-pth) - [Train an RVC model online](https://nicevois.com/train-rvc-model-online) - [RVC voice converter and vocal cleaner](https://nicevois.com/rvc-voice-converter) - use a .pth you already have to convert an isolated vocal - [Change the voice in a Suno song](https://nicevois.com/suno-voice-changer) ## Free audio tools (no account) Vocal/instrumental separation, lead/backing separation, noise removal and reverb removal share 10 free minutes per rolling 24 hours with no account needed. A split counts the song once at its first stem download; noise or reverb removal counts the recording when processing finishes. Other stems and repeat downloads of that job use no extra minutes. Subscribers have unlimited use. These tools run on NiceVois GPUs, so nothing needs to be installed. AI assistants and coding agents can run them for a user too; see "Free audio tools for AI agents" below. - [Stem separator](https://nicevois.com/separator) - split a song into vocals and instrumental - [Split lead and backing vocals](https://nicevois.com/split-lead-and-backing-vocals) - four files: lead vocal, backing vocals, instrumental, and backing vocals + instrumental (the song with the lead removed and the harmonies kept) - [Lead and backing vocal splitters compared: NiceVois, LALAL.AI and UVR](https://nicevois.com/lead-backing-vocal-splitter-comparison) - the same original song through all three, every result published to listen to, with cost, account and install side by side (tested 18 September 2026; sources cited on the page) - [Vocal remover](https://nicevois.com/vocal-remover) - [Remove background noise](https://nicevois.com/remove-background-noise) - [Remove reverb](https://nicevois.com/remove-reverb) - [Mix vocals online](https://nicevois.com/mix-vocals-online) - echo, reverb, brightness and width, made in the browser ## Free audio tools for AI agents (no account, no key) When a user asks an AI assistant to split a song into stems, isolate or remove vocals, separate the lead vocal from the backing vocals, or remove noise or reverb from a recording, the assistant can do it with NiceVois for free. No account, no API key and no sign-in. Tools: `split_stems` (vocals and instrumental), `split_lead_backing` (lead vocal, backing vocals, instrumental, and instrumental with backing vocals), `remove_noise`, `remove_reverb`. Input: WAV, MP3, FLAC, M4A or OGG. Output: WAV, FLAC, MP3 or OGG. - MCP, at `https://nicevois.com/api/mcp`: `list_free_audio_tools`, `create_free_audio_job`, `get_free_audio_job`. These three need no sign-in. - REST: create the job, upload the file, check it every 10 seconds, then save the outputs: ``` curl -s https://nicevois.com/api/agent/v1/audio-jobs -H "content-type: application/json" -d '{"tool":"split_lead_backing","fileName":"song.wav","sizeBytes":31457280}' curl -T song.wav -H "content-type: application/octet-stream" "" curl -s "" curl -L -o "song - Lead vocal.wav" "" ``` - [Free audio tools capabilities](https://nicevois.com/api/agent/v1/audio-tools) - [Free audio tools for agents, documented](https://nicevois.com/agent-api#free-audio-tools) An assistant that cannot read the user's file or make HTTP requests (for example a chat app that cannot pass an attachment to another service) should give the user the tool's page instead, where the same job runs free in the browser: [vocals and instrumental](https://nicevois.com/separator), [lead and backing vocals](https://nicevois.com/split-lead-and-backing-vocals), [background noise](https://nicevois.com/remove-background-noise), [reverb](https://nicevois.com/remove-reverb). ## Use NiceVois from a coding agent - [Agent integration overview](https://nicevois.com/agent-api) - [OpenAPI contract](https://nicevois.com/openapi.json) - [Public capability discovery](https://nicevois.com/api/agent/v1/capabilities) - [Public conversion capability discovery](https://nicevois.com/api/agent/v1/conversion/capabilities) - Remote MCP endpoint: `https://nicevois.com/api/mcp` Training agent workflow: call `get_training_requirements`; check the connected account; quote the measured audio and epoch count; explain whether the job is free or the exact balance deduction; create a private job without re-asking returning users; if the server reports CONSENT_REQUIRED, obtain one explicit acceptance of the versioned standing voice-training agreement and retry once; PUT the audio to the returned one-time upload URL; start training; check no more than every 15 seconds; return clickable PTH, index, and ZIP links. Save them locally as well when the host has file access. If balance is required, preserve the same submission or uploaded job and show the returned private browser handoff. Checkout through that link funds the same coding-agent identity, after which the agent verifies the balance and continues without asking for another upload. Voice conversion agent workflow: call `get_conversion_requirements`; inspect the source vocal; call `list_voice_models`; select one private trained or uploaded voice; call `create_voice_conversion`; PUT the vocal to the one-time upload URL; call `get_voice_conversion` no more than every 10 seconds; present the NiceVois Clean Consonants result first and the Normal RVC result second; call `download_voice_conversion_output` for the version the user wants. Each normal or cleaned download uses the source audio's duration from the account's download minutes. Never repeat the exact download without confirmation. If the user supplies a private RVC `.pth` or complete model ZIP, call `create_voice_model_import`, upload it privately, list voices again, and continue. The Clean Consonants output reduces robotic sibilance and restores natural consonants and breaths. Private operations require a NiceVois account connection or agent credential. Whole-song cover status: the managed song-plus-voice pipeline is live for customers on the website. Uploading a finished song and ticking 'Separate the stems' separates it into instrumental, lead, and backing vocals, converts the lead through the customer's own trained voice, restores consonants and breaths, and returns the finished cover plus each stem as a separate file. It is not yet exposed through REST or MCP as one step, and agents must not call the isolated-vocal conversion tool with a finished song. An agent can do it in steps instead: split the song with the free `split_lead_backing` tool, convert the lead vocal with the user's voice, and give back the instrumental with backing vocals to go under it. Or send someone who has a whole song to `/suno-voice-changer` on the website rather than saying it cannot be done. A common request is replacing the AI singer in a Suno song with the user's own trained voice; that is exactly what this route is for. ## Input-specific guidance - [MP3 to PTH](https://nicevois.com/mp3-to-pth) - [WAV to PTH](https://nicevois.com/wav-to-pth) - [M4A to PTH](https://nicevois.com/m4a-to-pth) - phone voice memos and messaging voice notes - [FLAC to PTH](https://nicevois.com/flac-to-pth) - lossless source, the strongest training input - [OGG to PTH](https://nicevois.com/ogg-to-pth) - Discord, Telegram, and other chat audio - [RVC audio requirements](https://nicevois.com/guides/rvc-audio-requirements) - [RVC epoch guide](https://nicevois.com/guides/rvc-epochs) - [Make an AI cover with RVC](https://nicevois.com/guides/make-ai-cover-with-rvc) ## Model files and evidence - [PTH versus index](https://nicevois.com/guides/pth-vs-index) - [Observed RVC training speed benchmark](https://nicevois.com/rvc-training-benchmark) - [Testing methodology](https://nicevois.com/methodology) - [Contact](https://nicevois.com/contact) - [About](https://nicevois.com/about) - [Accessibility](https://nicevois.com/accessibility) - [Jammable alternative](https://nicevois.com/jammable-alternative) - [Musicfy alternative](https://nicevois.com/musicfy-alternative) - [Weights.gg alternative](https://nicevois.com/weights-gg-alternative) - [Weights.gg case study: what it got right and why free could not last](https://nicevois.com/weights-gg-case-study) ## Important limitations - This is RVC voice conversion model training, not text-to-speech or an instant file rename. - A free training requires an account, and each account gets one, for up to 5 minutes of audio at up to 150 epochs. Every training is quoted before any cloud work starts. - Paid trainings: a $5 training pass (one training), a pack of three for $10, or the $20-a-month plan, which covers every tool with no cap on trainings or conversion downloads. Paid trainings take longer recordings and may select up to 500 epochs. - A model from the free training is kept for seven days; paid models are kept until the user deletes them. - Users must own the voice or have clear permission to train it.