Can you train dictation software to your accent?

Usually not in the way the word suggests. The dictation tools we checked do not let you retrain the model they use. Some offer a smaller step: a voice profile that the app learns from a short recording, or a list of words it should expect. Research shows that adapting a model to one speaker can reduce errors on accented speech a great deal. Those methods are mostly research tools, though, so check your app's settings before you assume it can learn your voice.

Adapting a model to one speaker has cut errors in research

The gains from personalization are large in the studies. In a 2007 study of foreign-accented English, the best combination of pronunciation modelling and acoustic adaptation cut word error rate by 46 to 71 percent relative ¹. That was an older research system. A 2019 Google study of accented speech reported a 35 percent relative reduction in word error rate after personalization ².

These are research methods, and the consumer tools we checked do not expose them. Expect a smaller gain from a consumer app than the best study reports.

Some apps ask for a voice profile, and it does less than the name suggests

The most common version you will meet is a voice profile. On Windows, Microsoft says Voice Isolation needs a quick one-time setup so that Voice access can learn to recognize your voice ³. Voice Isolation's main job is to filter out other speakers and background noise, so the profile helps the system focus on you. It is not an accent model. Microsoft also says that deleting your voice recording removes your voice signature from the device ⁴, which is worth knowing before you set it up.

Some tools skip training altogether. Nuance's training guide says Dragon Professional Anywhere needs no speech profile training, so it is ready to use straight away ⁵. Training was not always optional. IBM's Tangora research system, an early voice-activated typewriter, had to be trained by each individual user ⁶. Some current tools work without a profile.

Check your privacy settings before you give an app your voice

Some apps keep audio to improve their models, so check the setting before you turn on training. On a Mac, Apple's Improve Siri and Dictation setting allows the company to store and review a sample of your Siri and Dictation audio ⁷. Read the privacy setting for each app before you turn on training, and decide what you are willing to share.

What to do: test the change, not the feature

1. Find the options. Look in the app's dictation or speech settings for a voice profile, a custom word list or an option to improve recognition. If none is listed, the app probably does not learn from your voice.
2. Record a baseline. Dictate a passage of about 200 words and count the errors. Divide by the number of words to get your word error rate. Keep the passage and the result.
3. Add your words first. A custom list is the cheapest change, and it helps with names and technical terms. Apple's Voice Control accepts a text file with one term on each line ⁸. Keep the list short, and add a word only when the tool gets it wrong more than once.
4. Repeat the passage. Run the profile setup, then dictate the same passage again in the same room with the same microphone. Compare the two results.
5. Keep what works. A profile that lowers the error rate on your passage is worth keeping. One that does not is not, whatever the setup screen promised.

If a tool has no profile and no word list, the regional setting and a headset microphone are the next steps. The question on why dictation does not understand your accent explains the other causes, and the question on which speech to text works best with an Indian accent covers the choice of tool.

Where the evidence is thin

The studies above use a research system, a small set of speakers, or a disordered-speech population. The results show that adaptation can work. They do not tell you how much a consumer app will improve on your voice. Measure it yourself, and assume a smaller gain than the best studies report.

Our guide to speech recognition and accents covers the wider research, and the languages overview lists the other guides in this series. For a broader English plan, see English voice typing for non-native speakers.