Public SaaS
Create an account at voxis.nyata.ai, buy credits in rupiah, and use the full Voxis workspace right away. We run and maintain the service.
Mixed-language transcription that can run on your own servers.
Voxis labels each speaker and prepares a summary and to-dos from a recording or a link. Our public service stores your files encrypted in Switzerland.
Free to start · no card · no bot in your meetings
Tested on 15 public recordings · 15.4 hours · 6 languages and dialects · every word open to your ears
- Speaker 201:13
Oke, kita akan mulai ya. Do you want some chocolate?
- Speaker 301:16
You can start. You want to open it?
- Speaker 201:18
How do you tolerate his weakness slash kekurangan dari peanut sorbet?…
Every recording belongs to its creator. Recordings play from their creators’ own YouTube channels; the transcripts and reference texts are published here so you can check them. Nothing was edited. If you own one of these recordings and want it removed, write to us.
15 public recordings. Every word open to your ears.
We ran 15 podcasts, court hearings, talk shows, business interviews and a financial news programme through Voxis exactly as a customer would: Indonesian, Singlish, Malaysian English, Taiwanese Mandarin, Spanish and English, often inside one sentence. Play any moment, read what Voxis wrote, and judge for yourself. The mistakes are there too.
- 73 minAWARE survey & the kebaya row, Yah Lah But #857
- 71 minMixed in one sentenceDeddy Corbuzier × dr. Tirta
- 30 minMixed in one sentenceCinta & Vasco Q&A
- 41 minMixed in one sentenceThe Spanglish Show, ep. 53
- 97 minMixed in one sentenceLeyendas Legendarias #391
- 58 minMixed in one sentenceLazy Homegirls Brown Bag Podcast, ep. 201
- 25 minKasus 3 tahun Bandung, Nerror
- 68 minMixed in one sentenceSemua dituduh buzzer
- 99 minMixed in one sentenceThe KK Show 332
- 29 minProf Hibnu & forensic expert on the Sutrimo exhumation
- 43 minDon't start a company just because you want to IPO
- 44 minNvidia deepens chip ties with MediaTek
- 90 minRay Dalio: I predicted the 2008 crash
- 2 more recordings
Word error rate
4.7%on recordings in one language13.9%where speakers mixed languages
Across all 15: word error rate 4.7% on single-language recordings and 13.9% where speakers mixed languages; 75.2% of switched phrases kept in their original language. References were built by an independent model and cross-checked, not by ear; treat a point or two as noise.
Where Voxis got it wrong
Record it, upload it, or paste a link.
48:12
Cinta & Vasco Q&AIDEN
Speaker 2 dan Speaker 3 mengadakan sesi tanya jawab untuk menjawab pertanyaan warganet dari Instagram.
Terkait mitigasi skenario kiamat zombi, Speaker 3 menyarankan untuk segera keluar dari Jakarta menuju pulau seperti Kepulauan Seribu atau Bali, sementara Speaker 2 memilih pulau tropis yang kaya sumber daya seperti Pulau Rote atau Lombok.
However the audio reaches Voxis, you get the same things back: a transcript with every speaker named, four ready-made briefings — summary, key points, to-dos, and Q&A — and a clean export to PDF or Word. It works with the AI assistants your team already uses too, via MCP.
Your recordings stay yours.
Files are sealed the moment they arrive and opened only to do the work you asked for — never to train AI, never to mine.
Encrypted
Your content is protected with TLS in transit and AES-256-GCM at rest. Each organization has its own encryption key, held separately from its stored content.
Stored in Switzerland
Where does my data live, and who can read it?- Never used for AI training
- Your recordings are never used to train models and never mined. The transcription provider's copy is deleted as soon as the work is done.
- Delete means delete
- Deleting removes the audio and makes its transcript and summaries unavailable in Voxis. For a complete database wipe, contact us; encrypted backups expire on their normal schedule.
Use Voxis on our cloud. Or yours.
Start with the public service, or put Voxis inside infrastructure your organization controls.
Private deployment
We deploy Voxis to infrastructure your organization controls. Your team keeps authority over the environment, access, and operating policy; we help plan and run the deployment.
Contact usDifferent tools are good at different jobs.
Otter, Fireflies, and Notta are capable transcription products. Voxis is built for professionals who need a dependable record of sensitive conversations, deliberate support for language switching, and more control over deployment.
| VoxisFor professional work where the record, language mix, and deployment boundary all matter. | OtterA mature meeting assistant for connected productivity workflows. | FirefliesBroad meeting automation with a large integration ecosystem. | NottaFlexible meeting capture, transcription, and translation across devices. | |
|---|---|---|---|---|
| Best suited to | Sensitive professional conversations, multilingual records, and organizations with deployment requirements. | Cloud meeting notes for teams using established meeting and productivity integrations. | Meeting automation, conversation intelligence, and integration-heavy workflows. | General meeting notes, mobile transcription, and translation workflows. |
| Languages in one conversation | 26 languages supported, with up to five mixed in a single conversation. Select the expected languages and Voxis follows the code-switching among them. | Six published transcription languages. One is normally selected per recording; French-English switching is the documented exception. | Automatic switching across 60+ eligible languages is available in beta on Business and Enterprise plans. | Choose two languages from its supported bilingual set before recording or upload. |
| Ways to transcribe | Record in the browser without a bot, upload a file, or paste a public media link in the same workspace. | Record without a bot or import a file. Its documentation does not support transcription from a pasted YouTube link. | Record without a bot or upload a file. Direct downloadable media URLs are supported through its API. | Record without a bot, upload a file, or import a public YouTube URL in its web app. |
| Deployment | Public SaaS or a private deployment in infrastructure your organization controls. | Managed cloud service; public documentation describes US-hosted storage. | Managed cloud service; Enterprise can use private storage, while public documentation describes US processing. | Managed cloud service, with a device-local Privacy Mode for paid desktop users. |
| Pricing | The public service includes 120 free minutes, then usage-based pricing per transcribed minute. Summaries are included; there is no per-seat subscription. | Free tier, then per-user subscriptions. | Free tier, then per-user subscriptions. | Free tier, then per-user subscriptions; bilingual transcription may require an add-on. |
Built for multilingual professional life.
Code-switching happens wherever multilingual people talk, but how visible it is varies: in Southeast Asia it is everyday speech, while in most English-speaking countries it stays far less obvious. Voxis treats it as a professional transcription requirement, not a regional niche.
Compared against public product documentation on 25 August 2026. Features and plans change. Test each service with representative recordings and confirm deployment requirements with the vendor.
Questions.
The things people ask us before they sign up, and how to record so the transcript follows.
Before you sign up
A minute of audio costs Rp 400, and one credit is exactly one minute — a 12-minute recording costs 12 credits, rounded up to the started minute. There are no subscriptions and no seat fees. Summaries are free, your first 120 minutes are on us, and we don't ask for a card to start.
On the public service: QRIS, or a transfer from any Indonesian bank. For a private deployment: bank transfer to Indonesia. We're looking at more payment providers.
Your organization's files stay in Switzerland, locked with an AES-256 key of its own, held apart from the files themselves. The servers run AMD SEV confidential computing, so memory stays encrypted while the work runs, and our cloud provider can't read what the machine is working on. Our admin tools don't expose the content of your recordings, transcripts or summaries. Our sign-in service and the specialist key-management service guarding that key run in Switzerland too. Turning speech into text happens outside it: files you upload and recordings you make go to a speech-recognition service in the European Union, and a public link you paste goes to a second one in the United States and Europe. Google's AI writes the summaries. The privacy policy names them all.
Nothing you record is ever used to train AI. Delete a recording and the audio can't be brought back: there is no undo, and we never restore deleted content from a backup. Older encrypted backups expire on their normal schedule. The encrypted transcript stays in our database until you ask us to wipe it, and we do that by hand on request.
A specialist speech-recognition provider in the European Union turns the files you upload and the recordings you make into text. It holds SOC 2 Type II and ISO 27001 certification, and we have its copy deleted the moment the work is done. A public link you paste goes to a second speech-recognition service instead, which runs in the United States and Europe. The privacy policy names both. Our logs record that a request happened, never a word of what was said. Audio from a live recording is erased automatically after five days.
Recording well
Over Bluetooth a microphone sounds like a phone call, and the transcript suffers. A USB cable carries the whole voice. This is true on laptops and phones alike.
The microphone built into your laptop or phone is fine at arm's length. Lay the phone face up on the table, take off a thick case, and keep it away from the air-conditioning.
Put a compact USB speakerphone in the middle of the table, with every seat within about 1.5 metres. The Jabra Speak2 55 and Poly Sync 40 are typical choices.
Use a full-size speakerphone or a conference video bar, plugged in by USB. Their microphones follow whoever is speaking. We have had good results with the Jabra Speak2 75 and Logitech conference kits. Above eight seats, pair two units or add the maker's wired extra microphones, as the Poly Sync 60 and EPOS Expand 80 allow.
No table microphone reaches across a hall. Ask the sound crew for a feed from their mixing desk into your laptop, by USB or audio cable. Then every podium and floor microphone goes straight into the recording. With no sound system, clip wireless microphones on the speakers instead.
Any USB-C microphone that needs no app will do: a speakerphone, a clip-on, or a wireless kit with two transmitters, such as a DJI Mic or Røde Wireless. The wireless kit covers a bigger table than any phone can. Plug it in before you open Voxis; an iPhone uses whatever was connected last, and a web page cannot switch it for you. On iPhone, set Mic Mode to Wide Spectrum. Voice Isolation keeps one voice and mutes the rest.
Product names are references, not recommendations. We have no ties to any of these makers.
Start with your next meeting.
Free to start. No card and no bot in your meetings.
Planning a private deployment, need help with the public service, or have a question? Send us a note.