Can speech to text handle Hinglish, Spanglish and other mixed languages?

Speech to text can transcribe mixed speech such as Hinglish (Hindi and English) or Spanglish (Spanish and English), but accuracy drops on the mixed parts. The built-in settings we checked ask you to pick one language at a time, and the research we found shows a higher error rate on code-switched speech than on single-language speech ¹. We found little research on Hinglish and Spanglish specifically, so test a tool on your own sentences before you depend on it.

Our guide to multilingual dictation covers the wider options, and the question on dictating in two languages covers what switching can and cannot do.

Mixing languages inside a sentence is the hardest case

Researchers call switching between languages within a conversation or a sentence code-switching. A 2025 study by Liu and colleagues reports that spontaneous switching within a sentence causes language confusion for code-switching recognizers, and that accent bias blurs the sound boundaries between languages ².

Coverage is also uneven. A 2025 review found that a small subset of languages receives most of the attention, and that the languages covered by datasets are very unequal ³. If your mix is not one of the well-studied pairs, expect less support and more editing.

Error rates on mixed speech are high

In the 2025 study of two conversational corpora, Mandarin-English and Spanish-English, Whisper's mixed error rate on code-switched utterances was about 39 percent ⁴. The same authors found that Whisper significantly altered and reduced the speakers' original code-switching during recognition ⁵. These figures come from a research test, and they do not give the error rate for your own speech.

For Hinglish specifically, a 2025 Hinglish corpus paper cites recent evaluations showing that ASR models have a 30 to 50 percent relative increase in word error rate on code-switched speech compared with monolingual input ¹. A relative increase is a ratio, not a number of percentage points, so read it as a sign of the gap rather than an exact rate.

The word error rate on mixed speech is the number to compare when you test a tool. Compare it with the same tool's rate on a single-language sample.

A model trained on the mix does much better

Training on mixed data narrows the gap. A 2023 method adapted Whisper-small to Mandarin-English code-switching and reached a 14.2 percent mixed error rate while training only 5.6 percent additional parameters ⁶. The same authors say that multilingual models such as Whisper often struggle with the code-switching setting ⁷.

That result comes from a research setup trained on its own corpus. It is not what a general app delivers on your speech, and we found no adapted Hinglish or Spanglish model that we could test here.

What the consumer tools document

We did not find a Google, Apple or Microsoft page that says its dictation follows a Hinglish or Spanglish mix. The Gboard help we read supports automatic detection in six languages, but it does not describe mixing two of them inside a sentence ⁸. Dedicated Hinglish speech models exist from third parties, but we did not verify their accuracy, so test any of them before you choose one.

What to do

1. Choose the language that carries most of each sentence, and dictate in that language. Accept some errors in the other language, or type those words yourself.
2. Dictate in blocks, one language per message or paragraph, rather than switching inside sentences. The guide to changing the dictation language lists the steps by device.
3. Add names and recurring loanwords to the tool's vocabulary list, if it has one. The question on dictating foreign names and words explains how.
4. Test with ten or so sentences that match your real speech, including mixed ones, and count the errors. Ten sentences will not give a precise rate, but they show whether a tool is usable.
5. Proofread every mixed sentence before you send or publish it.

For Hindi and other Indian languages on iPhone, see Hindi and other Indian languages on iPhone dictation.