They Answered but You Can’t Hear Them: Voice and Video Calls in Your Vibe Project
Calls hit real-time behavior, permissions, and hardware all at once: incoming calls, echo, weak-network degradation, and recording rules — a hands-on guide with a worked funnel.

One thing up front: this guide is only about live voice and video calls — how to make them, how to answer, what to do on a weak network, what you can do mid-call, and what is left after hanging up. The Voice Input guide is about turning speech into text, which is not a live call; the Live Chat Support guide is about text-based customer service, which is not a call either; the Read Receipts and Presence guide is about message states, which has nothing to do with call states. The example product in this article, “Skylark” (a language-exchange app), and the people in it, “Xiao Man” and “David,” are fictional examples used to make the points concrete — not a real product or real users.
Many vibe projects add calling from one very specific moment: users have been typing in chat for a while, it feels too slow, and they want to just say it. So the author adds a voice call button and a video call button and considers the job done. Then the complaints arrive: they answered but heard nothing, the video froze into a slideshow, the call rang forever unheard, and the microphone stayed on after hanging up. Calling is one of the hardest features in a product because it hits three problems at once: real-time behavior, permissions, and hardware. A text message arriving three seconds late bothers nobody; a call freezing for three seconds makes the other person want to hang up. Text needs no microphone permission; a call has to get past permissions first. Text works the same on every device; a call behaves completely differently on an old phone, a Bluetooth headset, and a desktop browser. Until those three problems are understood, a call button only invites complaints.
1. Three call shapes — pick one and do it well
Calls usually come in three shapes: 1-on-1 voice, 1-on-1 video, and group calls. Their difficulty climbs in that order, so do not try to ship all three at once. 1-on-1 voice is the simplest: the lowest bandwidth, the lowest device demands, and the starting point for most products. Its bar is two words: hear clearly. As long as both sides can hear each other, with no echo and no piercing squeal, a voice call passes. 1-on-1 video is a step harder: on top of sound, there are picture, camera switching, front and back cameras, portrait and landscape. Its bar is: the picture does not freeze into a slideshow, and lips match the voice. Group calls are the hardest: three or more people talking at once need mixing, the picture has to split into tiles, the active speaker has to be highlighted, and someone dropping out has to be handled gracefully. Do not start group calls before 1-on-1 is solid.
Pick the shape by what calls are for in your product. Language exchange, counseling, and remote interviews only need 1-on-1 voice — video adds pressure, since not everyone wants to show their face on a first call. A designer walking a client through a proposal, or a teacher working through a problem with a student, fits 1-on-1 video better, because the picture and screen sharing matter. Team meetings and community events are what need group calls. In our fictional example, “Skylark” started with 1-on-1 voice only, ran it for two months until complaints dropped to zero, and only then added video; video had to run clean before group voice rooms were tried. Each step waited for the previous step’s complaints to clear. That is not slow; it is the only correct pace for a calling feature.
One more choice people miss: should calls work in the browser, or only in the app? On the web, browser calls face two hurdles: whether the browser grants microphone permission, and whether the browser kills the call when it goes to the background. A mobile browser sent to the background drops calls easily. In an app the experience is far better, but the user has to install the app first. A reasonable middle path: voice calls on the web, video calls in the app, with a clear note next to the button — “video calls work in the app.” Do not let users start a video call in the browser, watch it stutter, and conclude your product is broken.
2. Incoming calls: how to ring without startling anyone
The incoming call screen is the first impression of the whole feature, and the easiest to get wrong. The call may arrive while the phone is locked, while the user is watching a video, or while they are in a meeting. The screen has to do three things. First, make it obvious who is calling — avatar, name, and your relationship, such as “language partner · 12 chats so far.” Never show just a phone number or a user ID. Second, make Answer and Decline big and far apart. The user may be walking or driving when the call comes; buttons that are small or close together get mis-tapped. Third, cap the ringing. If nobody answers in 30 seconds, hang up automatically and log a missed call. Do not let it ring for two minutes and annoy the other person and everyone around them.
What happens after a missed call matters more than the call itself. When the user opens the product they should see a clear missed-call record: who called, when, and a callback button. The callback belongs right next to the record — one tap to call back. Never turn a missed call into a cold system notification that leads nowhere when tapped. One more detail: when the other person is already on a call and you dial them, you should see “the other person is on a call” instead of ringing until timeout. That state has to come from live call status; do not leave the user waiting on a call that will never connect.
Do Not Disturb is a must-have for incoming calls. Users sleep, meet, and study, and do not want calls then. Keep the setting simple: one master switch, plus an option for “only take calls from mutual follows.” Calls during Do Not Disturb should not ring — just log a missed call for later. And there is the background case: if the user switched the app to the background or locked the phone, can the call still ring? The app needs system-level incoming-call push, or locked phones will never receive a call. That push permission needs its own explanation — “so we can alert you about incoming calls while your phone is locked” — not a vague “for notifications.”
3. Microphone, camera, and echo: the permission and hardware hurdles
For call permissions, timing matters more than wording. The worst moment is right when the user opens the product — a “Allow microphone access?” prompt before they know what you want gets refused on reflex. The best moment is when the user taps the call button: before the system dialog, show your own page saying, “This call needs the microphone so you and the other person can talk. Nothing will be recorded.” Then show the system prompt. Users who understand the purpose say yes far more often. If they refuse, do not keep popping dialogs; instead put a guide in the call screen — “Microphone is off. Turn it on in Settings” — that jumps to the system settings page. The guide should spell out the path, like “Settings → Skylark → Microphone,” not just “please turn it on in Settings,” which leaves the user hunting.
Echo is the most common hardware problem in voice calls: the other person hears their own voice played back, as if speaking in a cave. It happens when speaker sound is picked up by the microphone and sent back. There are three layers of fix. First, software echo cancellation, which most calling services include — just turn it on. Second, suggest headphones: they physically break the echo loop, and the call screen can gently note “Echo detected — headphones will sound better” when echo is found. Third, volume: do not default the speaker to full volume; 70% is enough. Howling is echo taken to the extreme — two devices too close together with the volume too high, sound feeding back into a sharp squeal. The product should automatically lower the speaker volume when howling is detected, instead of waiting for the user to find the volume key.
Camera and microphone switching should feel effortless too. In a video call, a user who wants to flip from front to back camera to show the view outside should find the switch at hand — one tap, no digging through settings. A broken microphone or a camera held by another app needs a clear message: “The microphone is in use by another app. Close it and try again.” Not “device error.” Users are not engineers; “device error” tells them nothing. Before a call starts, a quick device check helps: test whether the microphone hears anything and whether the camera shows a picture, before dialing — far better than discovering the other side cannot hear you after the call connects.
4. Weak networks: degrading beats holding on
Calls fear bad networks more than no network. With no network at all, the product only has to say honestly, “Network unavailable,” and users understand. With a network that stutters and drops, the product’s behavior is what shows. The right move is to degrade — and to degrade proactively, not after complaints. The usual order: in a video call, drop the picture quality first, from high to standard to smooth; if that is not enough, turn the video off and keep voice; if voice still breaks, offer to reconnect. When degrading, tell the user what happened: “The network is weak, so we switched to voice mode.” Do not silently kill the video and leave the user staring at a black screen. Degrading is the product deciding for the user, and decisions deserve an announcement.
Set the quality levels in advance; do not make users tune them mid-call. Three levels are enough: high, standard, smooth. Default to standard, not high. High looks great on a good network, but on an ordinary one it is the source of the freezing. Users can change the default in settings and switch levels mid-call, but do not make the switcher too prominent, or users will think switching levels fixes everything. Reconnection logic needs thought too: after a drop, retry automatically three times, and only then tell the user “Connection lost. Check your network and call again.” While reconnecting, show “Reconnecting…” — do not throw the user out of the call. What users fear most during reconnection is not knowing what is happening, and one line of status keeps many of them around.
One weak-network case gets missed often: the other person’s network is bad, and you hear them breaking up. The product should tell whose network is at fault and word the message accordingly. If it is mine, “Your connection is unstable.” If it is theirs, “The other person’s connection is unstable, please wait.” If it cannot tell, just say “The connection is unstable” — do not guess. Call quality data should be recorded in the background: average delay, packet loss, and freeze counts per call — for the product, not the user. Users do not need to see “180 ms delay”; they need to know whether the call was clear or not. Data is for the product to read, not for the user to read.
5. Mid-call controls: fewer and bigger
The call screen has two rules: few buttons, big buttons. During a call the user’s attention is on the other person, not on the interface. Only four buttons are essential: mute, hang up, switch camera, and speakerphone. Everything else — screen sharing, raising a hand, adding people, recording, beauty filters — belongs in “More,” not spread across the screen. The hang-up button should be the biggest, reddest, and easiest to reach. The mute state must be obvious at a glance: red button, a slash across the icon, and a line of text saying “You are muted. The other person cannot hear you.” Many people mute and forget they muted, talking for a while before realizing nobody heard a word. That one line saves a lot of people.
Screen sharing is an advanced video-call feature, and three questions come first. One, are you sharing the whole screen or one window? Sharing the whole screen exposes notifications and chat history, so default to a single window and confirm once before starting: “You are sharing your screen. The other person can see what is on it.” Two, does your voice keep working while sharing? It must — you talk while presenting; picture and voice are not either-or. Three, how do you get back? One tap on “Stop sharing” returns to the video — no hunting. The raise-hand feature is essential in group calls: with many people you cannot just jump in, so you raise your hand first and the host calls on you. The button should be obvious, and a raised hand should tint the speaker’s tile border so the host can see at a glance who is waiting.
Do not lose in-call text chat either. During a video call, users often want to send a link, a number, or an address — typing is more accurate than speaking. The call screen should keep a chat entry that opens a half-screen chat panel without covering the video, and the chat history should be saved with the call record so users can find the link afterward. One more detail: what happens when the phone locks mid-call? The call screen should prevent locking, or at least leave a “Return to call” entry on the lock screen. Coming back from the lock screen to find the call still there, with the buttons still there, matters more than anything.
6. After hanging up: records, feedback, and recording rules
After a call ends, the product should keep three things: a call record, a quality check, and any recording the user actively asked for. The record is simple: who called whom, when, how long, who hung up, voice or video. It sits in the chat history, and tapping it calls back. Durations should read like a human wrote them — “3 min 42 sec,” not “222 seconds.” Missed calls show in red, answered ones normally, declined ones marked “Declined.” Keep records for at least 90 days, so users can find an old call when they need it.
Quality feedback is a gold mine for improving the product, but it has to be light to collect. After hanging up, ask one small question — “Was this call clear?” — with two buttons, “Clear” and “Not clear.” Tapping “Not clear” asks one follow-up: “What was wrong?” with three options: couldn’t hear, video froze, kept dropping. No five-star ratings, no long surveys; the user has just finished a call and has no patience for forms. Match the feedback against the backend data: user says unclear and the data is bad — a real problem; user says clear but the data is bad — maybe low expectations; user says unclear but the data is good — maybe a device problem. Only by reading all three together can you find the cause.
Recording is the most sensitive part, and the compliance line has to be drawn clearly. Recording a call requires both sides to agree; one side agreeing while the other does not know is illegal in many places. The right flow: when recording starts, the other person gets “The other person wants to record this call. Do you agree?” Only after agreement does recording begin, and a “Recording” marker stays visible for the whole call. Where the file is stored, how long it is kept, and whether it can be forwarded should all be stated before recording starts. One more suggestion: do not offer call recording by default. Add it later, once the product is bigger and someone has checked the legal side. The compliance cost of recording outweighs its benefits by a lot.
7. From dialing to hanging up: a fictional funnel, balanced
The funnel below is fictional, included to make call connection rates countable. Suppose “Skylark” logs 100 calls in a month. Of those, 94 reach the other person as an incoming call — the other 6 vanish: push not enabled, Do Not Disturb hours, or the app uninstalled. Of the 94 incoming calls, 88 are answered — the other 6 time out unanswered or are declined. Of the 88 answered, 81 establish audio and video normally — the other 7 stall at the media step: microphone permission refused, camera held by another app, or echo so bad the other person hung up. Of the 81 established calls, 76 run to completion — the other 5 drop mid-call and never reconnect. 76 out of 100 is a 76% overall completion rate. The absolute values do not matter; the point is where the problem sits. If the arrival rate is low, the problem is push and Do Not Disturb. If the answer rate is low, the problem is the incoming-call screen and ring time. If media setup fails often, the problem is permission guidance and device checks. If calls drop mid-way, the problem is weak-network degradation and reconnection. Funnels exist to locate problems, not to show off.
When the answer rate is low, do not rush to add features — slice the data first. Cut calls by network, device, and time: are old phones answering far less? Do calls after 10 p.m. drop far more often? Is incoming-call arrival unusually low for users in one country? Only by slicing can you find the real problem. In our fictional example, Xiao Man and David call three times a week; once David answered from the subway, the signal kept cutting out, and the call dropped twice. After the product started dropping to voice mode automatically on weak networks, their calls never dropped again — that is an invented story, of course; the principle is real, so do not quote the story.
8. Before launch, check this list
One, is the shape decided — 1-on-1 voice, video, or group — and in what order will the rest arrive? Two, is the incoming-call screen clear — who is calling, are Answer and Decline big enough, does ringing stop after 30 seconds? Three, can missed calls be called back — does the record show the time and a callback button? Four, is the permission timing right — requested when the user taps Call, with a path to Settings if refused? Five, is there a plan for echo and howling — is echo cancellation on, and does the product suggest headphones when echo is detected? Six, does weak-network degradation work — does switching from video to voice tell the user, and does reconnection try three times before giving up? Seven, are the in-call buttons few and big enough — is the mute state obvious at a glance, and can the hang-up button be found? Eight, is recording compliant — both sides agree before recording, with a marker visible throughout? Nine, is the post-call feedback light enough — a two-button question users will actually tap? Ten, do call records last — can any call from the last 90 days be found and called back?
Tick all ten and the calling feature is ready to ship. One last honest word: the launch bar for calls is not “it connects” but “it feels good.” Connecting is only the passing grade; clear sound, no drops, no surprises, and a record you can find afterward — that is what makes users tap the call button a second time. Start with 1-on-1 voice, get incoming calls, permissions, weak networks, and hanging up solid, then add video, then groups. Calling is the hardest feature in a product, and also the easiest way to earn trust — one clear call is worth ten smooth chats. Build it slowly; do not try to do everything at once.
Comments (0)
Related articles
One page should stand up on a phone, a tablet, and a computer. This guide follows MDN's Responsive web design, Using media queries, and the viewport meta reference: fluid HTML, the viewport tag, fluid grids and max-width, images that never burst their container, media query syntax, breakpoints set by content, mobile first, responsive type, testing capability instead of width, and a pre-launch checklist.
localStorage only stores strings and has no indexes. IndexedDB is a real database in the browser: structured data, far larger capacity, indexes, and transactions. Following MDN's Using IndexedDB, IDBDatabase, and IDBObjectStore: the five-step routine, opening, upgrades, keys, indexes, transactions, add versus put, get, cursors, ranges, and multi-tab upgrades.
You already call functions every day; this guide names what you have been doing. Following MDN's Functions, Arrow function expressions, and Rest parameters: declarations versus expressions, parameters and arguments, default values, rest parameters, spread syntax, return values, arrow function shorthand, the this trap, callbacks, and scope.