Skip to navigation

Pin a Model Version

View as Markdown

By default, Dictate uses the alias dictate-v1. The alias moves to a newer snapshot after 30 days’ notice, and the text can change a little when it does. If your code parses the text (a regex on an amount, a date in a fixed form), pin a dated snapshot. A dated snapshot never changes, and it is kept for at least 90 days after notice of its retirement.

How dictate-v1-2026-11-05 differs from dictate-v1-2026-11-04:

  • Dates. When the speaker says “March fifth”, 04 writes “March 5th” and 05 writes “5 March”.
  • Amounts. For “five hundred fifty thousand rupees”, 04 writes “550,000 rupees” and 05 writes “₹5,50,000”.
  • Cost. 05 uses about 16% more input tokens than 04.

A snapshot fixes the speech recognition model, the finishing model and both prompts. The other settings are yours.

import os
from sarvamai import SarvamAI
client = SarvamAI(api_subscription_key=os.environ["SARVAM_API_KEY"])
with open("note.wav", "rb") as f:
res = client.dictate.transcribe(
file=f,
language_code="en-IN",
model="dictate-v1-2026-11-05",
)
print(res.model) # every result names its snapshot
print(res.text)

Good to know

  • Every result names its snapshot in model, for a clip, a job, and on the live socket (session.begin and dictation.final). Log it.
  • On the live socket, pass model when you connect. It is fixed for the session.
  • An unknown name returns 400, and the message lists the names you can use.
  • Even with a pinned snapshot, results vary a little between runs. Names and wording can change, so check the text before you depend on it.