Skip to main content
Also known as speech-to-text (STT), transcription is the process of converting speech to text. We have built this endpoint with strong support for African languages.

​
Request

The transcribe() function can be used to transcribe audio. Pass either a url or a content to the transcribe function.

​
Parameters

  • language (required) - Language code: en, yo, ha, ig, am
  • content (optional) - Audio file content (binary data)
  • url (optional) - URL to the audio file
  • model (optional) - STT model to use: mansa_v1, legacy, human. Defaults to legacy
  • special_words (optional) - Custom words to help with recognition accuracy
  • timestamp (optional) - Timestamp granularity: sentence, word, none. Defaults to none
You must provide either content or url, but not both.

​
STT Models

The model parameter allows you to choose different speech-to-text models:
  • legacy (default) - Standard transcription model
  • mansa_v1 - Enhanced model with better accuracy for African languages
  • human - High-quality model optimized for human speech patterns

​
Timestamp Options

The timestamp parameter controls the level of timing information returned:
  • none (default) - No timestamp information
  • sentence - Timestamps for each sentence
  • word - Timestamps for each individual word
Examples are provided below as a guide for you.

​
Best Practices for Use

  • You can provide either the content (file) or url (str), but do not provide both.
  • The maximum file size is 25MB, we will support larger sizes in the future.
  • We only support mp3, wav, m4a, and ogg file formats.
  • If you provide url, ensure that access to the file is not blocked by authentication.
  • When transcribing, you should use the language code (e.g. en, yo, ig) and not the full text.

​
Response

The response for speech generation is in bytes.
  • The Content-Type is application/json
  • A request_id is returned for issue resolution with our support team.
Below is an example of a response from the transcription endpoint. 
    {
      "request_id": "86095cea-77d5-45ba-a093-0f800ac2c7df",
      "text": "Báwo ni olólùfẹ́ mi?"
    }

​
Language Support

Our speech-to-text model supports the following languages:
  • Hausa: ha
  • Igbo: ig
  • Yoruba: yo
  • Amharic: am
  • English: en
More info on languages can be found on the Languages page

​
Examples - file

Python
import os
from spitch import Spitch

os.environ["SPITCH_API_KEY"] = "YOUR_API_KEY"
client = Spitch()

with open("new.wav", "rb") as f:
    response = client.speech.transcribe(
        language="yo",
        content=f.read(),
        model="mansa_v1",
        timestamp="sentence"
    )
print(f"Text: {response.text}")

​
Examples - url

Python
import os
from spitch import Spitch

os.environ["SPITCH_API_KEY"] = "YOUR_API_KEY"
client = Spitch()

response = client.speech.transcribe(
    language="yo",
    url="https://myfilelocation.com/file.mp3",
    model="human",
    special_words="Spitch API"
)
print(response.text)
⌘I