Trusted Voice Research Infrastructure for NGOs, Governments & Global Development Partners
Language capabilities

“Do you support my language” is six questions, not one.

A language can be one a respondent types in, one they speak, one we can transcribe, one we can translate, one we can analyse, and one we can publish in — and it can pass any of those while failing the next. So we record them separately, and we never add them up into a single number of languages supported.

The six questions

Each is answered on its own for every language in the registry.

Text

Can a respondent answer in writing in this language?

Answered per language, from the capability registry. Recorded separately from every other row here, because a language can pass one of these and fail the next.

Voice

Can a respondent answer by speaking in this language?

Answered per language, from the capability registry. Recorded separately from every other row here, because a language can pass one of these and fail the next.

Transcription

Can that speech be turned into text?

Answered per language, from the capability registry. Recorded separately from every other row here, because a language can pass one of these and fail the next.

Translation

Can it be translated for analysis?

Answered per language, from the capability registry. Recorded separately from every other row here, because a language can pass one of these and fail the next.

Analysis

Can free text in this language be analysed?

Answered per language, from the capability registry. Recorded separately from every other row here, because a language can pass one of these and fail the next.

Reporting

Can a deliverable be produced in this language?

Answered per language, from the capability registry. Recorded separately from every other row here, because a language can pass one of these and fail the next.

What each answer means

The distinction that matters most is between something we measured and something a supplier published. They are not the same claim and this page does not let them look alike.

Validated

VoiceInsights measured this and recorded the evidence.

Supported with review

VoiceInsights measured this and it works, with human review before results are published.

Voice capture only

Audio in this language can be recorded and stored. It is not transcribed or analysed — capturing speech is not understanding it.

In validation

A supplier publishes support for this. VoiceInsights has not yet measured it, so we do not claim it works.

Not available

No configured provider supports this, and VoiceInsights has not measured it. It is not offered.

The registry, language by language

Read live from the capability registry. A language is a working language when a deliverable can be produced in it — that is what separates the researcher’s language from the respondent’s, and the two are not interchangeable.

Loading the capability registry…

What we refuse to say

These three are prohibited outright, and the prohibition is enforced in the code that builds this page rather than in a style guide someone has to remember.

What this page will not tell you

  • No speech-recognition accuracy figure is published for any African language; none has been benchmarked.
  • The platform performs no automatic language detection. A declared language is what the caller supplied.
  • The number of registered languages is not a capability figure. Registration, measurement and product editions are three different counts.

The interface is a separate question again

Which languages your team can work in is not the same as which languages you can collect in. The workspace and this website are translated into English, Kiswahili, French, Portuguese and Arabic. Collection reaches far beyond that list — and analysis reaches less far than collection does. Three different answers, kept apart.