Keyterm Prompting

Keyterms help Saaras v4 recognize important names, places, brands, and technical terms that may occur in an audio file — for example:

1
["Sarvam", "New Delhi", "Vistaar"]

Keyterms bias recognition; they do not guarantee that a term will appear in the transcript.

Rules

REST API

REST requests use multipart/form-data. Send keyterms as a JSON-encoded array in one form field.

1
import os
2
from sarvamai import SarvamAI
3
4
client = SarvamAI(api_subscription_key=os.environ["SARVAM_API_KEY"])
5
response = client.speech_to_text.transcribe(
6
    file=open("audio.wav", "rb"),
7
    model="saaras:v4",
8
    mode="transcribe",
9
    language_code="hi-IN",
10
    keyterms=["Sarvam", "New Delhi", "Vistaar"],
11
)
12
print(response.transcript)

Omit language_code to use automatic language detection.

Batch API

For batch jobs, keyterms is a normal JSON array inside job_parameters.

1
job = client.speech_to_text_job.create_job(
2
    model="saaras:v4",
3
    mode="transcribe",
4
    language_code="hi-IN",
5
    with_diarization=True,
6
    keyterms=["Sarvam", "New Delhi", "Vistaar"],
7
)
8
job.upload_files(file_paths=["interview.wav"])
9
job.start()
10
job.wait_until_complete()
11
job.download_outputs(output_dir="./output")

After creating the job, use the returned job ID to upload the audio, start the job, poll its status, and download the output using the standard Batch API workflow.

Batch with diarization

Add with_diarization. The same keyterms are applied while transcribing every audio chunk in the job — the downloaded result includes the full transcript and, when requested, speaker-attributed entries under diarized_transcript.

1
job = client.speech_to_text_job.create_job(
2
    model="saaras:v4",
3
    mode="transcribe",
4
    language_code="hi-IN",
5
    with_diarization=True,
6
    with_timestamps=True,
7
    keyterms=["Sarvam", "New Delhi", "Vistaar"],
8
)

See Enable Speaker Diarization for the full diarization output format.

Choosing the transcription mode

The keyterms format is unchanged across all supported modes:

For example, use mode="codemix" instead of mode="transcribe" when a keyterm such as PhonePe must remain in Latin script.