You opened Waze this morning, heard the same default voice you've heard for the past three years, and wondered if you could finally swap it for something better — maybe your own voice, maybe a cloned celebrity, maybe just an accent that doesn't grate by mile twelve. The search for waze voice packs turns up a confusing mix of official menus, GitHub repositories, file-replacement hacks, and vague promises about AI voice cloning. Most of that information contradicts itself.
Here's what's actually true, drawn from Waze's own community documentation, vendor tutorials, and the structural reality of how the app handles audio. You'll get the supported paths, the unsupported ones, the technical reasons true voice cloning can't (yet) deploy into navigation, and the use cases where voice cloning genuinely works today.
Table of Contents
- What Waze Voice Packs Actually Are (and the Myth of "Custom" Voices)
- How to Change Your Waze Voice on Android and iPhone
- Recording Your Own Voice in Waze: How "Add a Voice" Actually Works
- The Unofficial Path: Community Voice Pack Repositories and File Replacement
- Why You Can't Drop an AI-Cloned Voice Into Waze
- Where Voice Cloning Actually Works Today: 6 Production-Ready Use Cases
- Choosing a Voice Cloning Platform: A Decision Matrix
What Waze Voice Packs Actually Are (and the Myth of "Custom" Voices)
A Waze voice pack is the audio bundle Waze plays during turn-by-turn navigation — direction calls ("turn left in 500 feet"), distance announcements, hazard report acknowledgments, and start-of-drive greetings. According to Ridester, Waze voices are "audio prompts in different languages, accents, and styles," and the catalog you see inside the app is the result of Waze curating that audio for each market.
Three distinct categories of waze voice packs show up in the wild, and conflating them is the source of most confusion online.
Official built-in voices are the professionally produced packs Waze ships natively, surfaced under Settings → Voice & sound → Waze Voice. They vary by language and accent and require nothing more than a tap to activate, per both the Murf.ai blog and Mygpstools.
Waze celebrity voices are limited-run packs Waze releases periodically — characters, athletes, actors. They appear directly inside the same Waze Voice menu when active, with no sideloading. Speechactors documents this rotation, and Ridester notes these packs are produced under Waze's own licensing pipeline.
User-recorded "custom" voices are the third category, and this is where the myth lives. The in-app "Add a voice" flow lets you record every navigation phrase manually. Waze then plays those recordings back during turn-by-turn navigation. Per Murf.ai's tutorial, the user records each prompt one at a time and Waze stores the clips.
That last point matters: custom waze voices are not AI voice cloning. They are voice-memo replacement. You record your actual voice saying fixed phrases, and Waze plays those exact clips. There's no model. No generative speech. No way for the system to produce a phrase you didn't record — including, critically, street names. This is fundamentally different from true voice cloning, which builds a generative model capable of saying any text, and from Text to Speech systems that produce dynamic speech from written input.
The other piece of the confusion is the Waze Voice Pack Repository that turns up in multiple blogs (Mygpstools, Ridester, Speechactors). This is a community-maintained, unofficial GitHub-hosted collection of celebrity and themed packs. Users can install them through mobile browser links that hand off to Waze. It works — for now. It is not officially curated by Waze, and the Waze Community Forum is explicit that the company doesn't sanction it.
What follows walks through the supported paths first (official voice selection, in-app recording), then the unsupported ones (repository installs, file-level replacement), then the structural reason AI voice cloning can't be deployed into navigation today, and finally where voice cloning actually delivers value right now — in content production, not turn-by-turn directions.
How to Change Your Waze Voice on Android and iPhone
This is the official, supported path. It works identically on iOS and Android, takes under 60 seconds, and exposes every voice currently available to your account — including any limited-time celebrity waze voice packs Waze has surfaced in your region. No file access. No desktop sync. No third-party tools.
Open Waze and tap the main menu icon. On current builds this is the magnifying glass or "My Waze" entry at the bottom of the screen. Older versions surface the menu through a hamburger icon in the top corner. Per the Murf.ai blog, this is the entry point on every supported version.
Tap Settings (gear icon). It sits in the menu drawer. On some builds the section is labeled "Settings" directly; older versions nest it under "My Waze," as Mygpstools documents. Either way, the gear icon is the marker.
Open "Voice & sound." This section houses both the navigation voice and the sound-effect toggles (chimes, alerts, hazard notifications). Ridester confirms this is the universal label across recent app versions.
Tap "Waze Voice." The list shows every voice installed and every voice available for download, grouped by language. Voices not yet downloaded show a download arrow next to the name; downloaded voices show a play button for previewing. The DelftStack tutorial walks through this list visually.
Select a voice to preview, then tap to set as active. Waze plays a short sample on the first tap. Tapping the same voice again confirms it as the active navigation voice. The change applies immediately — no app restart, no settings save, no confirmation dialog. Per Murf.ai, the new voice takes over starting with the next prompt.
(Optional) Search by language or accent. A search bar at the top of the Waze Voice list lets you filter by language, accent, or character name. When the catalog runs 30+ voices, this is faster than scrolling. DelftStack's walkthrough demonstrates the filter functionality.
Troubleshooting and notes. If a voice you expected doesn't appear, the most common cause is an outdated app — Waze rotates celebrity packs in and out, and limited-time voices vanish when the campaign ends. Update the app and reload the voice list. The menu path is identical on iOS and Android; there is no platform-specific divergence at the official-UI level, per Mygpstools. And per the Waze Community Forum, there is no other officially supported install path — anything that asks you to download files or visit external sites is operating outside Waze's sanctioned interface.
Recording Your Own Voice in Waze: How "Add a Voice" Actually Works
Waze includes a built-in "Add a voice" feature that lets you record your own audio for navigation prompts. This is the closest the app gets to custom waze voices inside its supported feature set, and it's the source of a lot of misunderstanding about what Waze can and can't do. Set your expectations now: it's not AI, it's not text-to-speech, and it requires patience. According to the Murf.ai blog, the feature exists as a structured recording workflow, and Ridester documents the end-user experience as laborious but functional.
- Where to find it. The "Add a voice" button sits inside the Waze Voice list (Settings → Voice & sound → Waze Voice), typically at the top or bottom depending on app version. Tapping it triggers a safety acknowledgment screen before the recorder opens, per Murf.ai. You can't proceed to the microphone interface without acknowledging the warning.
- The mandatory safety warning. Waze forces every user into a pre-recording acknowledgment screen because custom recording is safety-relevant — navigation clarity affects driving decisions. Mispronounced street names or unclear instructions can cause real confusion at intersections. The warning is Waze's built-in liability control, and Murf.ai's tutorial confirms it can't be bypassed. Tap through, then the recorder loads.
- The phrase categories you must record. Waze breaks navigation prompts into category groups including Start of drive, Distances, Instructions, Reports, and Other. Each category contains multiple individual phrases — "Turn left," "In 500 feet," "Police reported ahead," "Continue straight," and so on. You record each phrase one at a time, working through the categories in sequence. Both Murf.ai and Ridester describe this as the core friction point of the workflow.
- Time limits per phrase. Each recording has a strict time limit per individual prompt. This forces tight, short takes — long pauses or extended phrasing would garble navigation timing during actual driving. Plan for clear, clipped delivery, not natural conversational pacing. Per Ridester, this constraint is by design and not negotiable. Re-recording a prompt that ran long is faster than fighting the limit.
- The fallback behavior. Any prompt you skip or fail to record gets played in Waze's default voice during navigation. This creates a hybrid output — your voice for the prompts you recorded, the default voice for everything else. Both Murf and Ridester implicitly recommend recording every prompt to avoid jarring voice switches mid-route. A partial set sounds strange in practice; the voice swaps every few turns.
- Saving and activating. Once recorded, your custom voice appears as a new entry in the Voice recorder list inside Waze Voice. Select it like any other voice. You can re-record individual prompts later without redoing the entire set — useful when a particular phrase didn't land the first time. Per Murf.ai, the recordings persist until you delete the custom voice entry.
Waze's "custom voice" is voice memo replacement dressed up as personalization — it records your voice saying fixed phrases, not a model that can say anything new.
The reality check: this feature is functional but laborious. Expect 30–60 minutes to record a complete set if you want zero default-voice fallback. And critically, it doesn't generalize. Waze can't say new street names in your voice because there's no model behind the audio — only playback of what you recorded. That generalization problem is exactly what platforms like a Voice Cloning API solve in other contexts: produce arbitrary speech from a short voice sample. Waze just isn't a context where that technology can plug in, which the next two sections explain in detail.
The Unofficial Path: Community Voice Pack Repositories and File Replacement
Beyond Waze's official menu, a parallel ecosystem of community-maintained voice packs exists — usually hosted on GitHub-based "Waze Voice Pack Repository" pages referenced by Mygpstools, Ridester, and Speechactors. These packs are unofficial. The Waze Community Forum states bluntly that "you can't install any [voice packs] except those Waze offers." What follows describes how the unofficial methods actually work and where they break, because they do work — until they don't.
The browser-link repository install method
The simpler unofficial path uses a mobile browser handoff:
- On the phone where Waze is installed, open the repository page in a mobile browser.
- Tap the install link next to the desired pack.
- Waze opens automatically and registers the new voice in its catalog.
- Navigate to Settings → Voice & sound → Waze Voice and select the new pack from the list.
This method appears low-friction — it looks like the official flow once the handoff completes — but it depends on two things staying true long-term: the repository remaining online, and Waze's current build still honoring the install URL scheme the link uses. Neither is guaranteed. Repository links break. Install handlers get deprecated quietly in app updates. The workflow Mygpstools and Ridester document works today; whether it works six months from now is a question those sources can't answer.
The manual file-replacement method
This is the advanced approach documented in the Waze Community Forum thread. It bypasses every install handler and operates directly on Waze's internal file structure.
Android path. Voice packs live at /storage/emulated/0/waze/sound. Each voice has its own folder containing multiple .bin audio files keyed to specific prompts. The folder name acts as the voice identifier inside Waze — renaming a folder breaks recognition, per the forum documentation. Waze looks for specific folder names when populating its Voice menu, and a renamed folder simply disappears from the list.
The replacement trick. The workaround documented by power-users is to empty an existing voice folder (keeping the folder name intact), drop the new pack's .bin files inside, and let Waze play those files when the original voice is selected. You're hijacking the slot, not adding a new one. The voice in the menu still shows the original name, but the audio that plays is the replacement. Per the forum, this is the only file-level method that consistently survives app restarts.
iOS path. On iOS, the equivalent flow uses iTunes file sharing to access Waze's internal "sound" folder. Export the folder to desktop, replace the contents of a target voice folder with the new .bin files (folder name unchanged), and sync back. The folder-name rule applies identically. The forum thread documents this as a working but high-friction approach that requires a Mac or PC, a USB cable, and a tolerance for iTunes.
Both file methods are unsupported. Waze updates can wipe these files, restructure the sound directory, or reject substituted audio outright. The official answer from the community forum remains that only Waze-provided voices are sanctioned.
| Method | Voice source | Difficulty | Officially supported | Risk on update |
|---|---|---|---|---|
| Official UI selection | Built-in catalog | Trivial — 4 taps | Yes | None |
| In-app "Add a voice" | Your own recordings | Moderate — 30–60 min | Yes | None |
| Repository browser-link install | Community packs | Easy on mobile | No | Handler may break |
Manual .bin replacement (Android) | Downloaded .bin files | High — file access | No | Files may be wiped |
| Manual replacement via iTunes (iOS) | Downloaded .bin files | High — desktop sync | No | Files may be wiped |
The Waze catalog operates as a closed loop — repository installs and file swaps work today, but they're guests in someone else's house, and the locks can change without warning.
The structural takeaway: every supported path runs through the official catalog or the in-app recorder. Every other route — repository installs, .bin swaps — works at the user's risk and could vanish with the next release. There is no public Waze API for voice pack submission, no developer program for navigation TTS integration, and no sanctioned route for deploying an AI-cloned voice. This isn't a technical gap waiting to be filled. It's a deliberate product boundary tied to driver safety, voice licensing, and quality control. Which is exactly why the question "can I clone my voice and use it as my Waze navigation voice" has the answer it does.
Why You Can't Drop an AI-Cloned Voice Into Waze
This section answers the question lurking behind most searches for waze voice packs: can I clone my voice (or a celebrity's voice) and use it as my Waze navigation voice? The short answer is no, and the structural reason matters because it explains where voice cloning does work and where it doesn't.
Modern voice cloning platforms build a generative model from a short audio sample. DubSmart's Voice Cloning needs as little as 20 seconds of audio; ElevenLabs, Murf, and HeyGen operate on similar sample lengths. That model can then say any text in the cloned voice — new sentences, new languages, names that didn't exist in the training data. This is fundamentally different from Waze's playback system, which serves pre-recorded clips tied to specific navigation events. Per Murf.ai, Waze custom voices are recordings, not generated speech. The two technologies aren't competing approaches to the same problem; they solve different problems entirely.
Three structural blockers sit between AI voice cloning and Waze deployment.
First, no public TTS or voice-cloning API exists for Waze. The community forum confirms voice options live exclusively inside the app's Sound and Voice settings. There is no documented endpoint, no developer program, no integration partner pipeline for third-party voice generation. A Text to Speech API can produce dynamic speech for any application that accepts standard audio input, but Waze does not expose that input surface.
Second, the file format is fixed. Waze plays .bin audio files keyed to specific prompts, per the forum documentation. There is no mechanism to feed dynamic TTS into the navigation engine at runtime. Even if you stood up a server that streamed cloned speech on demand, Waze has no way to receive that stream and play it as a navigation prompt.
Third, prompt-level binding caps everything. Even if you generated every Waze prompt with a cloned voice externally — recorded the output, converted to .bin, dropped into the folder using the file-replacement method above — you'd still be limited to the prompt set Waze plays. Your cloned voice could say "turn left in 500 feet" because that phrase is in the prompt list. It could not say "turn left on Maple Avenue" because street names are dynamic and Waze pulls them from a separate pipeline. The dynamic content remains in the default voice regardless of how sophisticated your cloned audio is.
The licensing and safety dimension reinforces the closed architecture. The mandatory safety warning Waze shows before in-app custom recording reveals how seriously the company treats navigation audio. Letting arbitrary AI-generated voices into a safety-relevant feature would create liability around mispronounced street names, unclear instructions, and impersonation of public figures. Officially curated celebrity voices, per Speechactors, are licensed and produced under Waze's own pipeline rather than user-submitted. The closed ecosystem is partly a product decision and partly a risk decision — and both reinforce each other.
The productive reframe: AI voice cloning is exceptional for content production — videos, podcasts, e-learning narration, dubbed marketing assets — where the platform you publish to (YouTube, your LMS, your podcast host) treats the output as a standard audio or video file. The constraint isn't the voice cloning technology. The constraint is whether the target platform exposes a way to plug a custom voice in. Navigation apps don't. Video platforms do — natively, because they accept whatever audio track you upload. This is why voice cloning has exploded in AI Dubbing workflows but remains absent from navigation.
The limit on cloned voices in Waze isn't the AI — it's the door. Waze doesn't open one for custom audio, and that's a product decision, not a technical accident.
Where Voice Cloning Actually Works Today: 6 Production-Ready Use Cases
If you came here looking to clone your voice for Waze, the answer is no — but the same technology solves real problems in content production right now. The constraint everywhere is integration. Voice cloning works where the platform accepts your audio. Below are the use cases where the integration path is open today, and where the economics make sense.
- Multilingual YouTube dubbing. Clone your voice once from a 20-second sample, then dub your videos into 33 target languages while keeping your vocal identity intact. This matters for creators expanding from English-only audiences into Spanish, Hindi, Portuguese, French, Japanese, or any supported market — the dubbed audio replaces your original track in the export, and viewers hear your voice in their language. AI Dubbing workflows handle the timing and lip-sync constraints automatically.
- Podcast episode localization. Record an English episode, generate localized versions in your own cloned voice, and publish region-specific feeds. Listeners in non-English markets get your voice carrying the content, not a stranger's dub or an obvious AI narrator. The audio masters export as standard WAV or MP3, which every podcast host accepts without modification.
- E-learning narrator consistency. Course producers can clone a single narrator's voice and use it across hundreds of modules without re-booking studio time. New module added six months later when the original narrator is unavailable? Generated in the same voice, no continuity break for the learner. This solves the staffing problem that kills most large e-learning libraries — voice talent moves on, and the catalog starts to sound like a patchwork.
- Corporate training videos at scale. HR and L&D teams clone an internal presenter or executive once, then use Text to Speech to generate compliance updates, onboarding videos, and policy changes without re-recording sessions every quarter. The Voice Cloning API lets internal tooling generate these assets on demand as policies change.
- Commercial voiceover libraries. Record a brand voice once, then generate spot variations, A/B-tested ad copy, and regional adaptations on demand. The original talent gets royalty terms negotiated up front; production gets near-infinite flexibility. The AI Dubbing API handles regional adaptations programmatically when the campaign needs to ship across 10 markets in a week.
- Backup voice for content creators. Lose your voice to illness, travel, or scheduling conflicts, and a cloned model lets you ship scheduled episodes or videos without breaking your release cadence. Audience continuity preserved, sponsor commitments honored, schedule intact. This is the safety net that turns voice cloning from a novelty into operational infrastructure.
Every one of these works because the target platform — YouTube, Spotify, LMS systems, ad servers — accepts standard audio or video files. There's no API negotiation, no closed ecosystem, no .bin file structure to reverse-engineer. You generate the audio, you upload, it plays. That's the integration model voice cloning needs, and it's why navigation apps remain the frontier they are. The technology is ready. The deployment surface is what determines where it actually lands.
Choosing a Voice Cloning Platform: A Decision Matrix
If Waze isn't where you'll deploy cloned voices, the next question is which voice cloning platform fits your actual project. The honest answer depends on four variables: how much audio you have to train the clone, how many target languages you need, whether you need API access or just a dashboard, and how you pay (subscription, credits, or per-call). The matrix below scores the major options against four common user profiles. Use it as a starting filter, not a verdict — test outputs with your own sample before committing.
| Requirement | Multilingual YouTuber | Corporate Trainer | Podcast Producer | App Developer |
|---|---|---|---|---|
| Minimum training audio | 20 seconds | 20–60 seconds | 30–60 seconds | API-driven, flexible |
| Target language count | 30+ languages | 5–15 languages | 5–10 languages | Use-case dependent |
| Output format needed | Video with dubbed audio | MP4, MP3 for LMS | WAV, MP3 for hosts | JSON / streaming API |
| API access | Optional | Optional | Optional | Required |
| Pricing model fit | Credits with rollover | Subscription or credits | Pay-as-you-go credits | Per-call API pricing |
The Multilingual YouTuber cares about clone speed and language breadth above everything else. A 20-second clone with 33 target languages covers expansion into Spanish, Portuguese, Hindi, French, Japanese, and more without separate voice talent budgets. Credits with rollover matter because publishing schedules vary month to month — unused credits shouldn't expire when you take two weeks off. Compare against ElevenLabs (strong on voice fidelity, fewer target languages for full video dubbing) and HeyGen (video-first but priced higher per output). The decision usually comes down to language count and credit policy.
The Corporate Trainer prioritizes consistency over flexibility. They'll clone one narrator voice and use it for years across hundreds of modules. Subscription pricing makes sense when output is steady and predictable. Language count matters less here — most enterprises localize into 5–15 markets, not 30. Murf and DubSmart both fit this profile; choose based on integration with your LMS. Most LMS platforms accept MP4 or MP3 natively, and both platforms export both formats.
The Podcast Producer has the simplest profile: one voice, a few languages, episodic output. Pay-as-you-go credits beat subscriptions because production isn't continuous — episode cycles cluster, then gaps. WAV output matters for podcast hosts and editing suites that prefer lossless masters. Voice cloning here usually serves localization or backup-narrator use cases rather than primary production.
The App Developer lives inside the API. Dashboard quality is irrelevant; what matters is latency, voices-per-call cost, webhook reliability, and language coverage. This is where dedicated endpoints differentiate from dashboard-first products — the Voice Cloning API, the Text to Speech API, and the AI Dubbing API each address different integration patterns. Developers building voice features into apps want one of these three depending on whether the use case is identity preservation, dynamic content generation, or full localization pipelines.
Pick your persona from the matrix above. Then run this single test against any platform on your shortlist: record a 20-second sample of your voice in a quiet room (phone mic is fine), upload it, and generate the same 30-second test sentence in three target languages. Compare three things — how close the cloned output sounds to your original voice, how natural the foreign-language pronunciation is, and how long the generation takes from upload to playable output. That one test exposes more about real-world fit than any feature comparison sheet you'll read. If you're a YouTuber or content creator, start with the free tier — clone your voice, dub a 60-second clip, judge the output before committing credits to a full project. The platforms that survive that test are the ones worth keeping in your stack.