How do you transcribe Hindi or Hinglish audio to text?

You can transcribe Hindi with several tools, including Whisper-based apps that run on your own computer. The steps are the same as for any other language: set the spoken language to Hindi, transcribe the whole recording, and read the output against the audio. Hindi is recognized less accurately than English, and Hinglish, the everyday mix of Hindi and English, is harder still. For the wider picture, see our overview of transcription and the guide to transcribing audio in other languages.

Hindi error rates are higher than for read English, and the tool you pick changes them

On the FLEURS test of read sentences, Whisper large-v2 had a word error rate of 21.5 percent for Hindi ¹ That is about 5.1 times the rate Whisper large-v2 had for English on the same test ². Newer models score differently in Hindi, so the tool matters. A 2025 comparison on the same test gave these results.

The 21.5 percent is from the 2022 Whisper paper and tests large-v2. The table is from one 2025 paper and tests large-v3, so do not read the gap between the two Whisper scores as a version change. The 2025 table does show that a newer model is not automatically better in every language, and that Gemini had the lowest Hindi error rate of the three.

Read speech is the best case. On Mozilla's Common Voice, where volunteers record on their own devices, Whisper large-v3 scored 46.75 percent on Hindi ⁶, far above its FLEURS score. Your own recordings are more likely to resemble Common Voice than read sentences, so plan for the higher figure.

Hinglish is harder, and English words in the middle are where errors cluster

Hinglish mixes Hindi and English in one conversation. The figures above are for Hindi alone. We did not find a peer-reviewed, public Whisper error rate for Hinglish, so we do not give one.

The problem is structural. A 2025 study says spontaneous switching between languages within a sentence causes language confusion for code-switching recognizers ⁷. In practice, set Hindi as the language, then check the English words in the transcript first. Read the points where the speaker switches languages most closely, because that is where errors cluster. For the term, see code-switching.

Transcribe a Hindi recording on your own computer

Buzz is an app that transcribes and translates audio offline on a personal computer ⁸. It is a good fit for a recording you do not want to upload.

1. Choose Hindi as the spoken language before you start.
2. Transcribe the first minute or two and read it against the audio. If the output is in the wrong script or language, change the setting and run it again.
3. Transcribe the rest and check it at several points, not only at the start. Errors often appear when the speaker changes topic or speed.
4. Correct names, numbers and dates by hand, since these are the words a reader will rely on.

If you want English text too, the translation is a separate step. The question on translating a voice recording to English covers it.

Built-in dictation in Hindi depends on the platform

For speaking into a text box rather than transcribing a file, the options differ by platform. On iPhone and iPad, Apple's feature availability page lists Hindi (India) in its general Dictation section ⁹. The same page has a separate on-device list, and Hindi is not on it, so dictation may go to Apple's servers. Check the setting that shows whether your dictation is processed on the device.

For Windows and Android, we did not check the current Hindi lists in this review, so look in your own voice typing or keyboard language settings before you rely on them. Yap, which we make, is an on-device dictation app for Mac and Windows. For its current language list, see the download page.

What to do next

Set the language to Hindi, transcribe, and read the first minute against the audio. If the recording is mostly Hindi with some English, check the English words and the switch points. For a recording that matters, such as a legal or medical record, have a fluent speaker check it, or use a human transcriber.

For the other languages in this guide, see how to transcribe Japanese, Chinese or Korean audio and whether AI can transcribe African languages. For the error-rate term, see word error rate.