1. Getting Audio Into Your Broadcast Device

fluent.church runs entirely inside a standard web browser on your broadcasting device (laptop, desktop, tablet, or smartphone). To transmit clear speech, you must bring clean audio into that device:

• Dedicated soundboard feed (Recommended): Connect an auxiliary output (Aux Out), Matrix, or Subgroup feed from your mixing console (e.g. Behringer X32, Allen & Heath, Yamaha) into your computer using a USB connection to the soundboard, Dante Virtual Soundcard, or an USB audio interface or an XLR-to-USB cable.
• Isolate spoken vocals: Assign only speech microphones (pastor lapel, pulpit mic, headset or handhelds) to this aux bus. Do not route loud instruments, drum overheads, or choir accompaniment to this feed, as continuous music can confuse the speech-to-text models. Ensure the aux (or channels) are muted when no one is speaking.
• Device microphone (Testing only): You can select your device's built-in microphone for quick bench tests, but room reverberation and distance will degrade transcription accuracy during actual services.

Verifying the Live Audio Meter

In the Broadcast Console, look at the visual VU Meter:

• When a speaker is talking into the mic, the green meter bar must actively bounce and rise above the threshold marker.
• If the meter remains flat or stationary during speech, the browser has either selected the wrong input device or the system input gain is muted. Open your browser or operating system sound settings and ensure the correct USB interface is selected and unmuted.

2. Noise Gate: Threshold, Attack, Hold & Pre-Buffer Memory

fluent.church includes a client-side Voice Activity Detector (VAD) / Noise Gate running in a high-performance audio worklet. It automatically pauses audio streaming when nobody is speaking so you are not billed for silence.

How the Threshold Works

The Threshold slider (default around -42 dB) sets the minimum sound volume required to trigger speech detection:

• Threshold too low: Background sanctuary noise (air conditioning, organ idle hum, congregation rustling) will hold the gate open continuously, causing the system to broadcast silence and consume your pre-paid hours.
• Threshold too high: Quiet or soft-spoken speakers may be cut off because their voice does not cross the trigger line.
• Best practice: Have someone speak into the microphone at a normal preaching level. Adjust the slider so the red threshold line sits comfortably above room background noise, but well below normal speech peaks.

How the Attack Works

The Attack time requires incoming audio to stay above the threshold for a brief, continuous window (e.g. 300 ms) before opening the gate.

• This deliberate attack prevents transient, accidental noises—such as a microphone bump, a cough, dropping a piece of paper, or a door closing—from accidentally waking the gate and incurring broadcast billing.

Pre-Buffer Memory: Nothing Before the Attack Is Lost

Because the gate waits for the attack window to confirm genuine speech, you might wonder if the first consonant or syllable of the speaker's sentence gets clipped.

It is not lost. While the gate is muted, the audio worklet continuously records into a rolling circular pre-buffer (ring buffer) that remembers the preceding 1.2+ seconds of sound. The instant the attack criteria are met and the gate opens, this remembered pre-buffer is immediately transmitted to the transcription engine as an opening burst. The AI receives the entire sentence from the very first spoken sound without any clipped beginnings.

How the Hold Time Works

Natural human speech contains short pauses between words and sentences. The gate has an automatic Hold time (hangover): once speech volume drops below the threshold, the gate remains open for several seconds before closing. This prevents the audio stream from chopping off word endings, conversational pauses, or quiet breath intakes between sentences.

Default URL vs. Custom Domain

By default, your church receives a permanent link at:

https://fluent.church/@your-church-name

If you prefer a seamless branded experience for your congregation, you can link your own subdomain (e.g. captions.yourchurch.org or live.mychurch.com).

Setup Steps

1. Create a DNS CNAME record: Log in to your domain registrar or DNS host (Cloudflare, AWS Route 53, GoDaddy, Google Domains, etc.) and add a new CNAME record:
   • Type: CNAME
   • Name / Host: captions (or your chosen subdomain)
   • Target / Value: cname.fluent.church
2. Register domain in Admin Console: Log in to fluent.church, navigate to Admin Console → Custom Domain, enter your full subdomain (e.g. captions.yourchurch.org), and save.
3. Verify DNS: Click the Verify DNS button. The system queries public DNS to ensure the CNAME points to cname.fluent.church.
   • Note on DNS propagation: It is expected that global DNS propagation and verification can take up to 48 hours depending on your registrar's TTL settings.
   • Once verified, automated SSL certificates and secure WebSocket origin policies are activated immediately.

Usage-Based Pricing

fluent.church pricing starts at $1 per broadcast hour. There are no mandatory monthly subscriptions, setup fees, or cancellation penalties. You purchase pre-paid hours as needed.

• Credit Expiration: Pre-paid hours remain valid for 1 year from the date of purchase. If you purchase additional hours at any time, your entire balance is extended by another year.
• Free Starter Hours: Every new organization receives 4 free trial broadcast hours valid for 30 days to test in their sanctuary.

The 3-Stage Billing Lifecycle

Because real-time speech AI has fluctuating duration, billing follows an automated three-stage cycle:

1. Stage 1: Initial Reservation (Broadcast Start)
   When you start a live broadcast, a 1-hour credit hold is placed on your account to guarantee uninterrupted streaming.
2. Stage 2: Preliminary Adjustment (Minutes After Broadcast)
   Within a few minutes after ending your broadcast, our system reviews audio stream records and adjusts the reservation to reflect actual speaking duration and listener presence.
3. Stage 3: Final Settlement (Within 24 Hours)
   Within 24 hours, our backend reconciles audited billing tags from the underlying speech provider. Any unused fractions of hours are settled precisely to the second.

The Silence & Audience Guarantee (and Force Broadcast Override)

Under standard operation, you are never charged when there are no connected listeners, nor for periods of silence when the noise gate is closed. If your sound engineer accidentally leaves the console running after everyone has left, your broadcast hours will not drain away.

Force Broadcast (Bypass Gate) Override

The Broadcast Console includes an override switch: "Force Broadcast (Bypass Gate)".

• When to use it: If you need to test transcription before attendees arrive, if difficult room acoustics make noise gate tuning temporarily challenging, or if audience connectivity tracking fails to detect connected viewers, this override forces continuous audio transmission to the AI engine.
• Billing impact: Enabling this toggle bypasses both the noise gate and the zero-audience safety pause. Your account will be billed continuously for all time elapsed while this override remains active, regardless of room silence or listener presence. Always turn this toggle OFF for normal church services.

We Buy and Relay an External AI Service

fluent.church uses industry-leading speech and translation engines: Deepgram Nova-3 for real-time speech-to-text, and Meta M2M-100 and Llama via Cloudflare Workers AI for multilingual translation.

We cannot manually alter or fix AI transcription errors: Because these models run on external neural networks, fluent.church cannot manually train special dialect models, override phonetic interpretations, or fix individual mistranscribed words.

How to Maximize Live Accuracy

The accuracy of AI models is almost entirely determined by input audio quality:

• Use close-mic techniques: Lapel, headset, or handheld microphones positioned 2–4 inches from the speaker's mouth yield 95%+ accuracy. Room microphones or boundary mics picking up room reverb will cause errors.
• Set healthy gain: Ensure the microphone signal is strong without clipping into the red.
• Encourage deliberate speech: Rapid mumbling or speaking over other people makes transcription difficult for both human listeners and AI.

Team Member Invitations

Church tech directors can invite other operators, pastors, and volunteers to their organization:

1. Go to the Admin Console → Organization Members.
2. Enter the team member's email address and assign their role (Admin or Broadcaster).
3. The team member can immediately log in using that email via Google or Microsoft.

Add Multiple Admins to Prevent Lockouts

We strongly advise designating at least two administrators for your organization.

Because fluent.church relies on federated authentication and has no manual support desk, having only one registered administrator creates a single point of failure. If that individual changes jobs, departs the church, or loses access to their Google or Microsoft account, your organization could be locked out of administrative controls. Having a secondary admin ensures your church can always manage settings, purchase hours, and invite new operators.

Lost Logins & Password Resets

fluent.church uses Federated OpenID Connect (Google and Microsoft/Entra ID) for all authentication.

We do not store passwords: Because authentication is handled entirely by Google or Microsoft, fluent.church has no password database and cannot send password reset links. If you lose access to your account, recover your password directly through Google Account Recovery or Microsoft Password Reset.

Dedicated Sound Booth Hardware Authorization

In church sound booths, the broadcasting computer or tablet is often shared by multiple rotating volunteers. You do not need volunteers to sign into their personal Google or Microsoft accounts on a shared venue machine.

Instead, administrators can enroll the computer itself as an authorized Sound Booth Device.

How to Enroll a Booth Device

1. Log in to the Admin Console with an administrator account and scroll to Sound Booth Devices.
2. Click + Enroll Booth Device.
3. Provide a recognizable label (e.g., Main Sanctuary PC, Balcony Mac Mini, or Youth Room iPad).
4. (Optional) Restrict the device to a designated channel (e.g. @main), or choose All Channels to allow the operator to select any channel.
5. Click Generate Enrollment Link.
6. Open the generated link on the sound booth computer, or scan the QR code with the device's camera.
7. The device will automatically authenticate and open the Broadcast Console—no password or third-party sign-in required.

Permissions and Revocation

• Broadcaster-Only Access: Enrolled booth devices have dedicated broadcasting permissions. They cannot view billing details, access payment methods, edit organization settings, or invite/remove members.
• Local Disconnect: An operator at the booth can sign out at any time by clicking Disconnect in the top navigation header.
• Remote Revocation: If equipment is decommissioned, replaced, or stolen, an Admin can click Revoke next to the device in the Admin Console. The device's security token is invalidated immediately on the server.

Low Bandwidth Requirements

• Broadcaster: Streams lightweight Opus audio packets (under 64 kbps). Any reliable broadband, Wi-Fi, or cellular connection is sufficient.
• Attendees: Only receive tiny compressed text frames via WebSockets (under 2 kbps per user). Hundreds of attendees can read captions simultaneously without saturating your church Wi-Fi.

Preventing Screen Sleep (Wake Lock)

fluent.church automatically requests a Screen Wake Lock in modern browsers (Safari on iOS, Chrome on Android, etc.) to keep the phone screen active while captions are streaming. If an attendee's phone still locks, ensure their device is not on ultra-aggressive "Low Power / Battery Saver" mode, which disables web wake locks.

Multiple Microphones & Overlapping Speakers

When multiple people speak simultaneously, speech AI models prioritize the loudest voice or may interleave speech fragments. For the clearest captions, encourage speakers to avoid speaking over one another.

Latency and Delay

Real-time English transcription typically appears on audience screens in 300–700 milliseconds, which feels almost instantaneous to listeners reading along. Translated languages (e.g. Spanish, Ukrainian, French) have a typical latency of 1.5–2.5 seconds due to full-sentence linguistic context processing.

Displaying on Projectors or Live Streams (OBS / vMix)

You can project live captions directly in your sanctuary or overlay them onto your video live stream:

• Open https://fluent.church/@your-church-name in a browser window and press F11 (or enter full-screen mode).
• In OBS Studio or vMix, add a Browser Source pointing to your listener link with a custom transparent CSS background.

Induction Loops and Auracast

fluent.church captions are a supplement to, not a replacement of, induction loops or Auracast hearing aid assistive systems. They provide visual accessibility for individuals who do not use hearing aids, as well as real-time language translation.

fluent.church is operated by:

Tingdahl IT Management AB
Sweden

For billing discrepancies, legal notices, or confirmed system defect reports:

legal@tingdahl.it

Please note: As stated above, technical setup consulting and personalized AV configuration advice cannot be answered via email. Please refer to the sections above or consult your local church sound technician.