Frequently asked questions

Tõlk.fm is a live AI translation platform for events. Organisers create an event room, stream or upload source audio, choose one or more target languages, and share a QR code. Audience members scan the code, pick their preferred language, and listen to a real-time translated audio stream — no app download required.

Tõlk.fm is designed for event organisers running multilingual conferences, summits, workshops, tours, panel discussions, and corporate events. Anyone who needs to address an audience in more than one language simultaneously can use Tõlk.fm in place of, or alongside, human interpretation.

No. Organisers manage events through the Tõlk.fm web dashboard at tolk.fm. Audience members open the join link in any modern mobile browser — no app download, no login, no account needed.

Sign in to your Tõlk.fm account, click New event in the dashboard, choose your event mode (live translation, scheduled playback, or hybrid), select your source language and up to eight target languages, and save. You will receive a shareable join link and a QR code that you can display on screen or print.

There is no hard cap on simultaneous listeners per channel. Tõlk.fm uses a broadcast architecture: the AI generates one translated audio stream per target language and distributes it to all listeners on that channel. Adding more listeners does not increase translation cost.

Yes. You can run a test event at any time from the dashboard. For live translation, test in the same room and with the same microphone or mixer you plan to use at the real event. Pay attention to latency, audio quality, and language accuracy before going live. Credit usage accrues during test runs at the same rate as production events.

Always rehearse with your actual audio source. Microphone placement and room acoustics strongly affect translation quality.

In live translation mode, source audio is captured in real time — from a microphone, mixer, or other audio input — and streamed to the AI translation engine. The translated audio is broadcast to listeners within a few seconds. This mode is ideal for conferences, keynotes, and any event where the content is not known in advance.

In scheduled playback mode, the organiser uploads a pre-recorded audio or video file before the event. Tõlk.fm prepares translated audio tracks in all target languages. At the scheduled start time, listeners on each language channel hear the translated version of the recording in perfect sync.

Hybrid mode combines prepared translated tracks for pre-recorded segments with live AI translation for live segments such as Q&A, panel discussions, or speaker remarks.

Audience capture modes control how speaker audio is collected:

• Conference / broadcast — Audio is captured centrally through the organiser's mixer or microphone. This is the default and works for most events.
• Open floor — Audience members can request the microphone from their phone. One person speaks at a time. The organiser or a moderator approves requests.
• Hybrid panel — Panelists use dedicated speaker-monitor devices while the floor remains open to the audience.

Speaker monitor mode gives the event speaker or interpreter a dedicated view showing live translations as they are generated. The speaker can also approve or deny open-mic requests from the audience.

You can add or remove target languages during a live event from the runtime controls in the dashboard. You can also switch between translation engines (OpenAI and Gemini) while the event is running. Changing the fundamental event mode requires ending and recreating the event.

The available languages depend on the translation engine chosen for the event.

If your target language is not in the OpenAI list, switch to Gemini in the event settings to access the full catalogue.

Both engines provide real-time speech-to-speech translation, but they differ in several ways:

You can switch engines live during an event. Test both before your event to choose the one that best fits your audience and content style.

You can run up to eight target language channels simultaneously per event. Each active channel is billed independently.

Translation accuracy depends on audio quality, speaking pace, accents, technical vocabulary, and background noise. For most conference and presentation contexts, AI translation is clear and understandable. However, AI outputs can contain omissions, mistranslations, or unsuitable phrasing.

Do not use Tõlk.fm as the sole translation source for medical, legal, financial, safety, emergency, or other high-stakes decisions. Always provide human alternatives where accuracy is critical or legally required.

When using the OpenAI engine, you can select from a range of synthetic voices and adjust playback speed. When using Gemini, the engine preserves the speaker's original voice characteristics.

Tõlk.fm captures audio from the organiser's device microphone or from a connected external audio interface, USB mixer, or audio adapter. For large venues, connect a direct feed from the PA or mixing desk.

Live translation introduces a few seconds of latency. The AI engine needs to hear a phrase before it can translate it. Listeners should understand that translated audio will trail the speaker slightly.

AI translation is optimised for one speaker at a time. When multiple voices overlap, accuracy may drop. Use open floor or hybrid panel mode to give the floor to one speaker at a time.

For scheduled playback, you can upload standard audio and video formats. The source audio is extracted, translated, and prepared before the scheduled start time.

Listeners receive the translated audio channel they selected. Volume is controlled through their device.

Share the join link or QR code with your audience. Listeners open the link, choose their preferred language, and start listening — no account or app download required.

Yes. In event settings you can define a custom join slug — for example tolk.fm/my-conference. Custom slugs must be unique across all events.

Headphones are strongly recommended to prevent feedback loops that degrade translation quality for everyone.

There is no listener limit. All listeners on the same language channel receive the same broadcast stream.

With Open floor or Hybrid panel mode, listeners can request the microphone. The organiser or speaker approves the request and that audio is translated and broadcast.

The listener's browser will attempt to reconnect automatically. Refreshing the join page reconnects them to the live stream at the current position.

Tõlk.fm charges $0.09 per minute per active translation channel, with a minimum charge of $15.00 per event regardless of actual usage.

Go to the Billing page in your dashboard and choose a top-up plan or enter a custom amount (minimum $5). Payments are processed by Stripe. Funds appear in your wallet immediately after payment.

Your wallet is charged at the end of each event based on actual minutes used and active language channels, subject to the $15.00 per-event minimum.

Active translation channels will stop when the wallet balance reaches zero. Top up before each event with enough funds to cover the expected duration.

Cost = duration in minutes × number of target languages × $0.09, with a $15.00 minimum. Example: 90-minute talk with 3 languages = $24.30.

Refunds are provided where required by law or approved case-by-case. Contact billing@tolk.fm with any billing concerns.

Live audio is processed transiently for real-time translation and is not retained by default. Uploaded media for scheduled playback is stored until you delete it or close your workspace.

Yes — audio is sent to the AI provider you select (OpenAI or Google Gemini) for translation. We use business-tier APIs under which your data is not used to train their models.

No. Listeners join through a link or QR code without registering. Session and connectivity logs are collected to provide the stream and troubleshoot issues.

Tõlk.fm is operated by Pixibit OÜ, an Estonian company subject to GDPR. Organisers are responsible for ensuring their use complies with applicable local laws.

We recommend informing attendees:

• Audio from this event is being streamed to AI services to produce real-time translation.
• Translation is generated by an automated system and may not be perfectly accurate.
• Listeners who use the open-mic feature should be aware their voice is also being translated and broadcast.

Local laws in some jurisdictions require advance disclosure or consent before recording or streaming audience audio. Check your local requirements before the event.

Yes. The Tõlk.fm Public API is a REST interface for partner products to programmatically create translation events, launch live sessions, stop sessions, and retrieve public listener links. API keys are profile-scoped, managed in the Developers dashboard, and billed against your existing prepaid wallet. Documentation is at tolk.fm/developers/docs.

streamUrl is the public browser listener page for the live event (for example https://tolk.fm/your-code). For broadcast tools, use directStreamLinks from launch or POST /v1/events/{id}/direct-stream-links — WebSocket PCM16, HLS/AAC, or RTMP pull URLs on the media worker.

Yes. Route program audio into Tõlk.fm, share streamUrl with browser listeners, and mint directStreamLinks (WebSocket, HLS, RTMP) for OBS, vMix, or Castr. YouTube/Vimeo do not accept third-party translation tracks natively — run translated audio in parallel via HLS/RTMP or listener links.

First, check your internet connection and your wallet balance. If both are fine, try the following:

1. Refresh the organiser control room page.
2. Stop and restart the affected language channel.
3. Switch to the other translation engine (OpenAI ↔ Gemini) to see if the issue is provider-side.
4. Check the event error log at the bottom of the runtime view.

If the problem persists, email hello@tolk.fm with the event ID shown in the dashboard URL.

Audio instability is usually caused by one of:

• Weak listener Wi-Fi or mobile data — advise listeners to move closer to the Wi-Fi access point.
• Organiser's upload bandwidth — the audio capture device needs a stable connection to the Tõlk.fm media worker.
• Too many devices on the same Wi-Fi — ensure the event Wi-Fi is separate from the public guest network.

A few seconds of latency is normal for live AI translation. Very high latency (more than 10–15 seconds) may indicate high network latency, overloaded AI provider capacity, or an unusually slow audio stream.

• High network latency between the organiser's device and our media worker
• Overloaded AI provider capacity (rare)
• An unusually slow or noisy audio stream requiring more buffering

Try switching engines or restarting the channel. If the delay is consistently above 10 seconds, contact support with the event ID.

Make sure you have granted microphone permissions to the Tõlk.fm site in your browser. If you accidentally denied permission:

• In Chrome: click the lock icon in the address bar → Microphone → Allow.
• In Firefox: click the lock icon → Clear permissions → Reload and try again.
• In Safari: open Settings → Websites → Microphone → allow tolk.fm.

Your wallet needs to have at least the event minimum ($15.00) available before a live event can start. Go to the Billing page and top up your wallet, then return to the event and try starting again.

The event ID appears in the URL of the event control room page: tolk.fm/events/[event-id].