How many languages does Google Speech-to-Text support?

Google says its Cloud Speech-to-Text service supports over 125 languages ¹. That is a headline figure for a developer service. The supported-languages table is the exact reference, because it lists each language by code and by model. Find your language and region there, and confirm the model supports the features you need, before you plan around it.

Google's headline figure is over 125 languages

The product page gives "over 125 languages" as the figure for the service as a whole. The service is an API that software calls to turn audio into text. It is not a dictation app, so it does not appear in a text box on your phone or computer. If you want a dictation tool built on Google's speech technology for everyday use, see which dictation app supports the most languages.

The table lists codes, so its rows are not a count of languages

Google's supported-languages page tells developers to "Use only the language codes shown in the following table" ². Each entry pairs a language with a region or variant, such as Xhosa (South Africa) with the code xh-ZA. A single language often appears under several codes, so counting rows gives the number of variants, not the number of languages. If you need one figure, quote Google's own product-page number and name the model you mean.

Each model covers a different set of languages

The table is organized by model, with the features for each model listed in columns. For example, the table lists Xhosa (South Africa), code xh-ZA, for the chirp_2 model, with its features listed beside it ³. A language can be on the table and still lack a feature you need, such as automatic punctuation or word-level confidence scores. Check the columns for each model, not just the language name.

The headline number and the table can differ over time

Google adds models and languages as its service changes, so a figure on a product page can lag the table, and the table can change between your reading and your build. A number you copy into a plan will go out of date. Put the date of your check next to any figure you quote, and recheck the table before a launch or a purchase. If a language you need is missing from a model you want, ask Google or choose another model that covers it, rather than assuming it will work. The guide to dictation languages explains why a listed language is not always an accurate one.

What to do

1. Find your language in the supported-languages table and note the exact code, such as en-US.
2. Check the model column for the features you need, such as punctuation, before you choose the model.
3. Use the code exactly as the table writes it, as Google asks.
4. Test the model on a sample of your own audio. Support on the table is not the same as accuracy, and accuracy varies with accent and recording conditions. The guide to speech recognition accuracy by accent covers the accent side.
5. If your language is missing, check the table again before you assume it is unsupported. The table changes as models are added, so the page is the place to check.

For the other questions in this group, see what languages Dragon NaturallySpeaking supports and whether speech recognition is less accurate in other languages.