Number and language options

Overview

Scribe v3 adds two optional parameters to the batch Speech to Text API:

• numbers_format sets whether numbers in the transcript are written as digits (42) or as words (forty-two).
• secondary_language_codes lists other languages that may be spoken in the audio alongside language_code. This improves accuracy on audio where speakers switch between languages.

Both parameters are only supported by scribe_v3. Sending either one with another model returns an invalid_parameters error.

Format numbers

Set numbers_format to digits or words. When the parameter is omitted, the model chooses the format.

import os
from io import BytesIO
import requests
from dotenv import load_dotenv
from elevenlabs.client import ElevenLabs
load_dotenv()
elevenlabs = ElevenLabs(
    api_key=os.getenv("ELEVENLABS_API_KEY"),
)
audio_url = "https://storage.googleapis.com/eleven-public-cdn/audio/marketing/nicole.mp3"
audio_data = BytesIO(requests.get(audio_url).content)
transcription = elevenlabs.speech_to_text.convert(
    file=audio_data,
    model_id="scribe_v3",
    request_options={
        "additional_body_parameters": {
            "numbers_format": "digits",
        },
    },
)
print(transcription.text)

numbers_format controls how the model transcribes numbers. For other changes to how dates, times or units are written, use transcript editing.

Declare secondary languages

Set language_code to the main language of the audio and secondary_language_codes to the other languages you expect. Both accept ISO-639-1 or ISO-639-3 codes.

secondary_language_codes is a list. Repeat the form field once per language. A JSON array sent as a single string is not accepted.

The following example transcribes English audio that also contains Spanish and French:

transcription = elevenlabs.speech_to_text.convert(
    file=audio_data,
    model_id="scribe_v3",
    language_code="en",
    request_options={
        "additional_body_parameters": {
            "secondary_language_codes": ["es", "fr"],
        },
    },
)

Scribe v3 transcribes speech in the language it is spoken in. It does not translate between the main and secondary languages.

Combine with other parameters

Both parameters can be used together and with other batch parameters, such as keyterms, diarize and no_verbatim. See keyterm prompting to bias transcription towards specific terms.

Next steps