Skip to main content
​
Feature · SDKs

​
Official Python and JavaScript libraries

pip install namu-ai and npm install namu-ai: the whole API, typed responses, one error class per status, retries and timeouts built in. Open source under Apache 2.0 at namu-ai-lab/namu-sdks.
​
Feature · speech-recognition · /v1/transcribe

​
Speech recognition in Zarma and French

/v1/transcribe now takes language=dje (Zarma) and language=fr (French) as well as Hausa, served by a new recognition model. Transcripts come in lower case, without punctuation.
​
Change · all models · /v1/models

​
Models are named after their tasks

The models are now speech-recognition, speech-synthesis, translation and voice-agent (they were tatsuniya, tafiya, kora and air). The new names appear in the model field of responses and in /v1/models. Requests never named a model, so no code needs to change unless it reads that field.Every voice is now spoken by the same speech synthesis model.
​
Feature · speech-synthesis · /v1/speech

​
Zarma speech, and the voice mouhamad

/v1/speech speaks Zarma ("language": "dje") with the new voice mouhamad, who also speaks Hausa. The French voice fr is back.
​
Change · /v1

​
New address: api.namuai-lab.com

The API is served at https://api.namuai-lab.com. The earlier address keeps answering for existing integrations; please move to the new one.
​
Feature · speech-synthesis, translation · /v1/speech, /v1/translate

​
MP3 and Opus, and text translation

  • /v1/speech takes format: wav (default), mp3, opus or json. Opus is about a twelfth of the size of WAV.
  • /v1/translate translates text: send JSON with text, source and target.
​
Removed · /v1/speech, /v1/converse

​
The voice salim is withdrawn

We could not confirm the consent of the person behind the voice salim, so it was removed. Requests that name it return 400 unknown_voice (or unknown_persona); nothing falls back to another voice.
​
Launch · /v1

​
The Namu API opens in beta

API keys, speech synthesis, speech recognition, Hausa ⇄ French translation and the voice agent, with 30 free minutes of audio a month for every account.

​
How we change the API

  • Added fields and endpoints can arrive at any time. Write your code to ignore fields it does not know.
  • Changes that could break a working integration (removing or renaming a field, changing a default) are announced here, with the date, before they happen, except when we must act at once to protect people, as with a withdrawn voice.
  • Research models (the voice agent) can change without notice.