duration.ai
Duration · Leaning
Duration is a consumer technology lab that runs AI speech models directly on your phone, including the Crosstalk text-to-speech family and an on-device voice cloning app for iPhone and Android.
About duration.ai
What Is Duration
Duration is a consumer technology lab. Its bet is that inference cost, not model quality, decides which AI products ordinary people can actually use for free. Running models on the device itself pushes that marginal cost toward zero, so speech generation and voice tools work offline and never send your recordings to a server.
The company splits its work in two. The applied research side ports and shrinks open-weight models so they run fast on phone hardware, chiefly the Apple Neural Engine on iOS. That on-device inference is the core idea. The product side turns the research into consumer apps, starting with a flagship voice app that is already launched.
The clearest limit is hardware. The heavier models need recent phones, and Android support lags iOS because accelerators there are harder to use. That means older or budget Android devices get the lightest model only.
Getting Started
- Check whether your phone is supported. The flagship app ships its own model, and the smaller one runs from 2020 iPhones and 2021 Android phones upward.
- Install the flagship app on iOS or Android and open it.
- Record a short voice sample, roughly 5 to 10 seconds, in a reasonably quiet spot.
- Let the app clone your voice on the device, then generate speech in that voice and export or share the audio.
Product Information
A quick look at duration.ai's pricing, supported platforms, and performance.
Best for
The users, tasks, and scenarios where this tool fits best.
Users
- iPhone and Android owners who want AI speech tools without a subscription
- People who care about privacy
- Creators who need narration fast
Tasks
- Cloning a voice from a short sample
- Generating speech with accurate word timings
- Cleaning noisy voice recordings
Scenarios
- Recording in a messy room
- Building subtitles for short video
- Testing AI speech on older hardware
Key features
On-device text-to-speech
Duration builds the Crosstalk family of mobile AI models for text-to-speech, tuned for phones rather than servers. Speech streams faster than real time and starts almost instantly, and mobile chip acceleration keeps the load off the cloud. The tradeoff is that the best model stays on newer iPhones.
Zero-shot voice cloning
The models clone a voice from a short sample, and the flagship app does it entirely on the phone. You record 5 to 10 seconds of speech and get audio in your own voice right away. Nothing leaves the device, which matters if you would rather not upload your voice to a service.
Built-in word timings
Crosstalk models emit word timings as they generate speech, so subtitles line up without a separate transcription step. The timings are accurate enough to drive live, karaoke-style highlighting while the audio is produced. That removes a whole stage from most captioning pipelines.
On-device denoising
A ported denoiser cleans voice recordings on the phone, and Duration reports that it gets close to paid cloud tools on speech. It handles background music, a running fan and overlapping speakers, which makes a short clone sample far more usable. Reverb is the weak spot and stays in the audio.
Runs on older phones
The smaller model reaches back to a 2020 iPhone 12 Pro and a 2021 Fairphone 4, and it fits inside an app bundle with no separate download. Performance climbs with newer chips, and a recent flagship finishes a clip several times faster. This range is what makes on-device speech realistic for a wide audience.
Neural Engine acceleration
Duration tunes its iOS models for the Apple Neural Engine, which does this kind of math faster and at a fraction of the energy. That keeps the phone cooler and frees up resources for the rest of the app. Android still relies on the CPU because its accelerators are harder to use well.
Pros and cons
Pros
- Speech runs on the device, so the app works offline and keeps voice data private.
- The smaller TTS model runs on phones going back to 2020, not just current flagships.
- Built-in word timings cut a transcription step from captioning and subtitle work.
- Low time-to-first-audio makes generated speech feel instant rather than delayed.
Cons
- The highest-fidelity model is limited to recent iPhones, so Android and older phones get less.
- No published pricing or free tier details, so you cannot compare plans before installing.
- No public API, which rules it out for developers who want to call the models from their own apps.
- The denoiser leaves reverb in the voice, so heavily echoing recordings still need extra cleanup.
Frequently asked questions
Duration is a consumer technology lab that builds AI speech models and apps that run on your phone instead of the cloud. Its main product line is the Crosstalk text-to-speech family, plus a flagship voice app.
Related content
Explore related tools, skills, and articles for duration.ai.
duration.ai Alternatives
Coddy
Coddy · Coding · LeaningCoddy is a gamified platform to learn to code through short interactive lessons in 20+ languages, with a browser editor, an AI tutor and free certificates.
Marv
Cameron (independent developer) · Coding · Productivity · LeaningMarv is an AI cursor companion for Windows and Mac. It sees your screen, answers out loud, and draws arrows and highlights that show you the next click.
Teachoo
iAsk AI · Leaning · ChatbotTeachoo is a free AI homework helper that walks students through math, science, writing and coding problems one step at a time, with photo input, hints and flashcards.