Getting your Scam With Your Friends caller languages configured correctly is the difference between a phone bank full of sharp, accented marks and a room of silent, broken scripts. Non-English callers only speak once the right Qwen3 voice pack finishes downloading, so this guide walks through the whole pipeline.
Choosing a Caller Language and Installing TTS Voices
Every caller in the game is backed by a text-to-speech engine that reads their dialogue lines aloud during a scam call. The default install ships with a small set of English voices, but the roster expands dramatically once you pull additional language packs. Each caller profile stores a language tag, a voice model reference, and a personality script, and the game matches those three pieces before a single word hits your speakers.
The catch is that voice models are not bundled with the base download. They stream in on demand, which is why so many players see a caller appear in the list but hear nothing when the phone connects. Understanding that separation — profile versus voice asset — is the foundation for everything else in this guide.
Here is how the core pieces fit together:
| Component | What It Stores | Where It Lives |
|---|---|---|
| Caller profile | Name, personality, scam type | Local save data |
| Language tag | Locale code such as es, fr, ja | Caller config file |
| Voice model | Qwen3 TTS weights for that locale | Voice cache folder |
| Dialogue script | Line-by-line text the caller speaks | Caller JSON |
When all four line up, a non-English caller greets you in their native tongue with proper pacing and inflection. When the voice model is missing, the game either falls back to a robotic placeholder or mutes the caller entirely. The voice cache is the single most common failure point, so treat it as the first thing you inspect whenever a caller goes quiet.
The good news is that the system is forgiving once the assets are in place. You do not need to reinstall the game or edit save files — you just need the correct voice pack sitting in the right folder with a matching language tag. The next sections break down exactly how to get there.
Downloading Qwen3 Voices for Non-English Callers
The Qwen3 voice download is where most setups succeed or stall. The game requests a model file the first time a caller with a new language tag tries to speak, and that request can hang if your connection drops or the cache directory is locked. Players who watch the progress indicator freeze at a low percentage are almost always hitting a stalled download rather than a corrupt install.
A clean download follows a predictable sequence. Start the game, trigger a call from a caller whose language you have not used yet, and let the voice pack fetch in the background. Do not alt-tab aggressively or close the phone UI mid-download, because that interrupts the stream and leaves a partial file behind.
| Symptom | Likely Cause | Fix |
|---|---|---|
| Progress stuck under 20% | Interrupted stream | Delete partial file, retry |
| Caller speaks in English | Wrong language tag | Recheck caller config |
| No audio at all | Missing voice model | Force re-download |
| Crackling or cutoffs | Low-quality fallback voice | Install high quality pack |
If you keep seeing the bar freeze, the underlying problem may be broader than a single voice file. The Qwen3 loading troubleshooting guide covers the deeper cache and memory issues that cause repeated stalls, including the voice download stuck behavior that frustrates so many players.
Once a download completes, the model is cached locally and future calls in that language load instantly. That is why the first call in a new locale feels slow while every call after it is snappy. Budget a few minutes of patience for each new language you unlock, and the payoff is a permanently faster roster.
Verifying a Completed Voice Download
After the progress bar fills, confirm the asset actually landed. Open the voice cache folder and check that the model file size matches the expected range for that locale — a file that is only a few kilobytes is a failed download wearing a success message. Relaunching the game and placing a test call is the fastest real-world verification, since a working caller will speak a full sentence without gaps.
If the test call still produces silence, the language tag in the caller profile probably does not match the model you downloaded. Cross-check the locale code, because a mismatched tag forces the engine to hunt for a voice that does not exist and quietly fail.
Choosing High Quality Voices Over Default Packs
Not all Qwen3 voices are created equal. The game offers a compressed default pack that downloads fast but sounds thin, plus a high-quality pack that takes longer to fetch and delivers noticeably richer audio. For non-English callers especially, the quality gap is obvious — accented dialogue in a compressed voice can slur consonants and make scam lines hard to follow.
The trade-off is storage and bandwidth. High quality packs are several times larger, so if you are juggling many languages you will feel the footprint. A sensible approach is to install high quality voices for the two or three languages you use most and leave the rest on the default pack.
| Voice Pack | Download Size | Audio Quality | Best For |
|---|---|---|---|
| Default | Small | Thin, robotic | Quick testing |
| High quality | Large | Rich, natural | Main languages |
| Fallback | Minimal | Flat, monotone | Emergency only |
Players who care about immersion tend to standardize on high quality for their primary locale and accept the default elsewhere. The difference shows up most in long dialogue trees, where a natural voice keeps the scam believable and a flat one breaks the illusion mid-call.
Matching Voice Quality to Caller Personality
Voice pack and caller personality interact more than most players expect. A fast-talking con artist reads better with a crisp high quality model, while a slow, menacing caller can survive on the default pack without losing impact. If you are building custom callers, pairing the right voice pack with the right personality script is what makes a character feel intentional rather than random.
Fixing Muted or Silent Text-to-Speech Callers
A caller who connects but says nothing is the most common complaint around caller languages, and it almost always traces back to the text-to-speech pipeline rather than the caller profile itself. The engine needs three things to produce sound: a valid language tag, a cached voice model, and an unlocked audio output. Break any one of them and the caller goes mute.
Work through the failure points in order. First, confirm the caller's language tag is set to a real locale the game supports. Second, verify the voice model exists in the cache. Third, check that no other application has grabbed exclusive audio control.
-
Language tag mismatch — the caller requests a locale with no matching model, so the engine silently skips audio.
-
Partial voice file — a stalled download left a corrupt model that fails to load without an error message.
-
Audio device conflict — another app holds the output device, muting all in-game speech.
-
Memory pressure — the engine drops the voice model when RAM runs low, muting callers mid-session.
That last point catches people off guard. On longer sessions with many languages loaded, memory pressure can evict a voice model that was working fine an hour earlier. If callers go silent only after extended play, memory is the likely culprit rather than a broken download.
When Typed Dialogue Becomes the Workaround
If voice output refuses to cooperate, the game lets you fall back to typed dialogue so a call can still progress. It is not as atmospheric as hearing a caller speak, but it keeps the scam moving while you sort out the audio side. Players dealing with persistent speech recognition or input problems often lean on this fallback, and the voice recognition troubleshooting guide explains how typed input fits into the broader audio setup.
Building a Multilingual Caller Roster
Once individual languages work, the fun part is assembling a roster that spans several locales. A diverse phone bank keeps scam sessions fresh, because each caller brings a different accent, pacing, and cultural flavor to the con. The trick is managing downloads so you are not waiting on five voice packs at once.
Plan your roster around the languages you actually want to hear. Pick a primary locale for your main callers, add one or two secondary languages for variety, and keep the rest uninstalled until you need them. This keeps the voice cache lean and reduces the odds of a stalled download blocking a session.
| Roster Strategy | Languages Loaded | Download Load | Variety |
|---|---|---|---|
| Focused | 1-2 | Light | Low |
| Balanced | 3-4 | Moderate | Medium |
| Collector | 5+ | Heavy | High |
A balanced roster hits the sweet spot for most players. You get enough linguistic variety to keep calls interesting without drowning your storage or your patience in download queues. If you want to see which callers pair best with which languages, the best callers breakdown ranks the roster by how well each personality performs across locales.
Pairing Languages With Custom Callers
Custom callers let you assign any supported language to a personality you built yourself, which is where the system gets genuinely creative. You can craft a caller who switches languages mid-conversation, or build a character whose accent is central to their gimmick. The caller personality customization guide walks through the JSON structure that controls these assignments, including how language tags and voice references are stored per character.
Managing Voice Cache Storage and Performance
Every language you install adds weight to the voice cache, and that weight eventually shows up as slower load times and higher memory use. The engine keeps recently used models in memory for fast access, so a bloated cache full of languages you rarely touch competes for the same resources your active callers need.
Pruning is straightforward. Remove voice packs for languages you have not used in weeks, keep the ones tied to your core roster, and re-download on demand if a caller you retired suddenly becomes relevant again. Because models re-fetch automatically when a matching caller speaks, deleting a pack is low-risk — the worst case is one slow first call.
-
Audit monthly — check which language packs you actually used in the last few sessions.
-
Keep core locales — never prune the languages tied to your main callers.
-
Clear partial files — delete any voice model under a few kilobytes, since those are failed downloads.
-
Watch memory during long sessions — if callers mute late in a session, the cache is too heavy.
Performance-minded players who run many languages simultaneously should also keep an eye on overall memory usage, since a heavy voice cache stacks on top of the engine's baseline footprint. Trimming unused packs is one of the cheapest performance wins available, and it costs you nothing but a re-download if you change your mind.
Rebuilding a Corrupted Cache
If the cache gets corrupted — usually from a crash mid-download — the cleanest fix is to wipe the voice folder entirely and let the game rebuild it. You will re-download every language you use, but you start from a known-good state instead of chasing individual broken files. This nuclear option resolves the stubborn cases where one specific caller refuses to speak no matter what you try.
Frequently Asked Questions
Why do my non-English callers stay silent after connecting?
Silent non-English callers almost always mean the Qwen3 voice model for that locale never finished downloading. Check the voice cache for a partial or missing file, then trigger another call in that language to restart the fetch. A complete model plus a matching language tag restores audio immediately.
How long should a voice download take?
Most single-language packs finish in a few minutes on a stable connection, while high-quality packs take longer because they are several times larger. If progress sits under 20 percent for more than a few minutes, the stream has stalled and the partial file needs to be deleted before retrying.
Can I use high quality voices for every language at once?
You can, but the storage and memory cost adds up fast. Most players install high quality voices for their two or three main languages and leave rare locales on the default pack. That balance keeps audio rich where it matters without bloating the cache.
Do custom callers support any language?
Custom callers can use any locale the game ships a voice model for, assigned through the caller's JSON config. If you tag a caller with a language that has no installed model, the engine fails silently, so always confirm the voice pack is present before assigning an exotic locale.
What causes callers to go mute mid-session?
Memory pressure is the usual cause. When RAM runs low, the engine evicts voice models to free space, which mutes callers that were speaking fine earlier. Pruning unused language packs and closing background apps keeps the cache light enough to stay resident through long sessions.
Getting caller languages right comes down to three habits: keep voice models fully downloaded, match language tags to installed packs, and prune the cache so memory stays free for the callers you actually use. Do that, and your multilingual phone bank will speak clearly in every accent you throw at it.