Skip to main content
Multilingual voice feedback

Let people answer in the language they speak.

For some open questions, a short spoken reply is easier than typing on a phone. HeySpeak keeps the original recording and creates a transcript in one of 13 supported languages.

Test a multilingual prompt
Test your language and audio conditions before a larger rollout

What is multilingual voice feedback?

Multilingual voice feedback lets someone answer an open question by speaking instead of filling a long text field. HeySpeak records the reply, transcribes supported languages, and keeps the original audio available when wording, tone, or transcription accuracy needs checking.

Current coverage

13 transcription languages

HeySpeak uses the Voxtral Mini Transcribe 2 model through Mistral’s voxtral-mini-latest endpoint. Mistral published this language list in February 2026.

Check Mistral’s current model coverage
  • English
  • Chinese
  • Hindi
  • Spanish
  • Arabic
  • French
  • Portuguese
  • Russian
  • German
  • Japanese
  • Korean
  • Italian
  • Dutch

What the evidence says

Speech was faster for Mandarin in one controlled study

A Stanford, Baidu, and University of Washington study compared speech recognition with mobile keyboards for short English and Mandarin message transcription. In its quiet lab setting, Mandarin speech entry was 2.87 times faster than the Pinyin keyboard. The study involved 32 participants and measured an upper-bound task, so it does not prove that every Chinese speaker prefers voice or that the result applies to Japanese, Korean, Arabic, or Hindi.

Read the mobile speech and typing study

Voice is worth testing when

  • The audience already sends voice notes in the channel where you reach them.
  • The question invites a story, explanation, reaction, or example.
  • People are answering on a phone and can choose a private moment to speak.

Text is usually better when

  • The answer is a rating, date, address, number, or another structured field.
  • The person is in a noisy, public, or privacy-sensitive setting.
  • Exact spelling matters more than tone, context, or speed.

What HeySpeak does today

The product supports multilingual transcription. It is not yet a fully localized survey platform or automatic translation layer.

Question
You can write it in the audience’s language.
Receiver interface
English or German today.
Transcript
Created in a supported spoken language.
Summary
Stays in the transcript’s language.
Source
The original audio remains available for review.

See it in a real workflow

Language is one part of the fit. The audience and question still matter more.

Creators

Podcast listener questions

Collect spoken questions or stories for one planned episode.

Broadcast

Radio audience prompts

Ask one question and review replies before a later show.

Guide

Voice input at work

Decide when a spoken reply fits and when written input is safer.

Common questions

Which languages can HeySpeak transcribe?
HeySpeak currently uses Mistral's Voxtral Mini Transcribe 2 model. Mistral lists 13 supported languages: English, Chinese, Hindi, Spanish, Arabic, French, Portuguese, Russian, German, Japanese, Korean, Italian, and Dutch.
Does HeySpeak support Hebrew or Thai transcription?
Not in the current supported-language list for the transcription model HeySpeak uses. Do not rely on HeySpeak for Hebrew, Thai, or another unlisted language without testing a representative recording first.
Can I get an English summary of a Chinese or Japanese answer?
Not automatically. HeySpeak currently writes the summary in the same language as the transcript. You can copy the text into a translation workflow afterward, but HeySpeak does not offer a summary-language setting today.
Is voice always easier than typing in a non-Latin script?
No. That depends on the person, language, setting, keyboard, and question. A controlled study found faster speech entry than Pinyin keyboard entry for short Mandarin messages, but that result should not be generalized to every language or real-world situation.
What does the person answering see?
They see the question and response controls in a supported mobile or desktop browser. The receiver interface is currently available in English and German. You can write the question itself in another supported language, but the surrounding interface is not yet localized into all 13 transcription languages.
What if the transcript gets a name or phrase wrong?
Transcription quality varies with language, accent, microphone, background noise, and specialist vocabulary. The original recording stays available, so you can check the source when the transcript looks uncertain.

Test one language before inviting a larger audience.

Record a representative sample with the accent, device, and background noise you expect in the real campaign.

Test a multilingual prompt