Migrating Speech-to-Text from Gemini to Sarvam

This guide covers migrating Gemini audio transcription calls to Sarvam. If you’re building a full voice agent with live, bidirectional audio rather than transcribing recorded audio one call at a time, start from the LiveKit or Pipecat voice agent build guides instead: they aren’t a 1:1 API swap and aren’t covered here.

Quick start

Base URL and auth header change the same way for every Gemini migration, so they’re folded in here rather than in a separate page:

With that, replace the transcription call:

1
import base64
2
from google import genai
3
4
client = genai.Client(api_key="YOUR_GEMINI_API_KEY")
5
6
with open("audio.mp3", "rb") as f:
7
    audio_bytes = f.read()
8
9
interaction = client.interactions.create(
10
    model="gemini-3.5-flash",
11
    input=[
12
        {"type": "text", "text": "Generate a transcript of the speech."},
13
        {
14
            "type": "audio",
15
            "data": base64.b64encode(audio_bytes).decode("utf-8"),
16
            "mime_type": "audio/mp3",
17
        },
18
    ],
19
)
20
21
print(interaction.output_text)

The rest of this page covers the full endpoint and parameter mapping, long audio, and common mistakes.

Endpoint map

Parameter concept map

Long audio and diarization

For files over 20 MB, upload first instead of inlining the audio.

1
from google import genai
2
3
client = genai.Client(api_key="YOUR_GEMINI_API_KEY")
4
5
uploaded_file = client.files.upload(file="long-meeting.mp3")
6
7
interaction = client.interactions.create(
8
    model="gemini-3.5-flash",
9
    input=[
10
        {"type": "text", "text": "Generate a transcript of the speech."},
11
        {
12
            "type": "audio",
13
            "uri": uploaded_file.uri,
14
            "mime_type": uploaded_file.mime_type,
15
        },
16
    ],
17
)
18
19
print(interaction.output_text)

Move diarized or long-form audio to the Batch API.

1
from google import genai
2
3
client = genai.Client(api_key="YOUR_GEMINI_API_KEY")
4
5
# Not available as a dedicated mode: Gemini has no speaker-labeling parameter.
6
# You'd need to prompt the model to attempt speaker labeling in its reply,
7
# with no guarantee of a consistent, structured format back.
8
uploaded_file = client.files.upload(file="meeting.mp3")
9
10
interaction = client.interactions.create(
11
    model="gemini-3.5-flash",
12
    input=[
13
        {"type": "text", "text": "Transcribe this and label each speaker."},
14
        {
15
            "type": "audio",
16
            "uri": uploaded_file.uri,
17
            "mime_type": uploaded_file.mime_type,
18
        },
19
    ],
20
)

See the Batch API guide for webhook setup and its Speaker Diarization section for the current per-job speaker limit.

What’s different by design

• A dedicated transcription endpoint. POST /speech-to-text and mode="transcribe" are purpose-built for turning audio into text, rather than a general model interpreting a natural-language request.
• A guaranteed-clean transcript field. response.transcript always holds just the transcript, with no prompt-engineering needed to keep a general-purpose model from adding commentary or reformatting the text.
• Flexible output modes on one model. saaras:v3 also supports translate, verbatim, translit, and codemix modes as explicit parameters, rather than needing a differently worded prompt for each.
• Diarization comes standard, not prompted for. Sarvam’s Batch API identifies and labels speakers with with_diarization and num_speakers, returning a structured diarized_transcript, rather than relying on a model’s best-effort attempt inside free-text output.
• Indian language depth is Sarvam’s specific strength. Saaras v3 covers 22 Indian languages, including Assamese, Konkani, Kashmiri, Sindhi, Sanskrit, Santali, Manipuri, Bodo, Maithili, and Dogri, with native handling for code-mixed speech like Hinglish.

Full example

A small wrapper function, before and after, showing the complete set of changes together:

1
import base64
2
from google import genai
3
4
client = genai.Client(api_key="YOUR_GEMINI_API_KEY")
5
6
def transcribe(path: str) -> str:
7
    with open(path, "rb") as f:
8
        audio_bytes = f.read()
9
    interaction = client.interactions.create(
10
        model="gemini-3.5-flash",
11
        input=[
12
            {"type": "text", "text": "Generate a transcript of the speech."},
13
            {
14
                "type": "audio",
15
                "data": base64.b64encode(audio_bytes).decode("utf-8"),
16
                "mime_type": "audio/mp3",
17
            },
18
        ],
19
    )
20
    return interaction.output_text

Common mistakes

See also

• Migration overview: base URL, auth, and SDK client changes
• Migrating Chat Completion
• Migrating Text-to-Speech
• Which Speech-to-Text API to Use
• Batch API
• Streaming API
• Errors & Troubleshooting