Skip to main content

Voice models

On-device voice typing needs a voice model on your phone. You choose which one; Loquela shows the models your phone can run and filters out the rest before you download anything.

Choosing a model​

Start with the language you speak. Then decide what matters more:

  • Quick final text: a focused model can be smaller and faster.
  • Sentence context: a contextual or multilingual model may handle names, technical terms and sound-alike words better.

Each model shows the same practical details in plain language: language support, download size, an estimate of the memory it needs, how quickly text appears, and its strengths and limitations. Sort by recommended order, download size or language coverage.

Download size is exact. Memory use and speed are estimates: they vary with the phone, the recording length and how warm the phone is. A model marked as possibly unreliable on your phone can still be downloaded, with a warning.

The full catalog, with the same details, is also on getloquela.com/models.

Free and Full models​

Free includes a curated set of on-device models. Full opens every compatible model in the current catalog, and lets you install verified custom model packages over HTTPS (each checked with SHA-256 before it is installed).

Downloads​

Models download through Android's download service, so a download can continue in the background and resume after an interruption. Loquela verifies each package before installing it. You need enough free storage for both the download and the extracted model.

Installed models stay listed above the recommendations, and you can delete a model to free space. When an update for a model you use is available, Loquela shows it.

Testing a model on your phone​

The best test is your own speech: try names, technical terms, punctuation, sound-alike words, long phrases, pauses and background noise, and check how long it takes from when you stop speaking until the text appears. Try it through each Bluetooth microphone you use as well.