Dubbing — Sarvam AI

Live in-chat dubbing → sarvam-mcp (sarvam_tools_dub). This skill = SDK code.

Translating a PDF/Word/Excel file's text? → translate/document — a different API.

[!IMPORTANT] Auth: api-subscription-key header — NOT Authorization: Bearer. Base URL: https://api.sarvam.ai. client.dubbing.upload() requires sarvamai>=0.1.31a1.

What this API does

Localizes a video or audio file into up to 12 Indian languages in one job, preserving each speaker's voice via voice cloning, and exports dubbed video, isolated audio, and SRT subtitles. Async, like translate/document, but exports fire automatically once the pipeline finishes — no manual per-language trigger call needed (unlike document translation).

Every integration follows five steps: create → upload → start → poll status → poll exports.

Quick Start (Python)

import os, time
from pathlib import Path
from sarvamai import SarvamAI

client = SarvamAI(api_subscription_key=os.environ["SARVAM_API_KEY"])
media = Path("sample.mp4")

# 1. Create job (does not accept the file itself)
created = client.dubbing.create(
    source_language_code="en-IN",
    target_language_codes=["hi-IN", "ta-IN"],
    export_options=["video", "srt"],
    voice_cloning=True,
    num_speakers=1,
    job_name=media.name,
)
job_id = created.data.job_id

# 2. Upload — the SDK handles the signed PUT + x-ms-blob-type header for you
client.dubbing.upload(created.data.upload_url, media)

# 3. Start the pipeline — wrap in try/except (see Gotchas: start() currently
#    throws a pydantic ValidationError even on success in some SDK versions)
try:
    client.dubbing.start(job_id=job_id)
except Exception:
    pass  # job starts server-side regardless — confirm via get_live_status below

# 4. Poll until the dub itself is done (exports still render after this)
TERMINAL = {"completed", "partial_failure", "failed", "deleted"}
while True:
    job = client.dubbing.get_live_status(job_id=job_id).data
    if job.status in TERMINAL:
        break
    time.sleep(15)

if job.status == "failed":
    raise RuntimeError(job.error_message or "job failed")

# 5. Poll export-status for the actual download links (limit defaults to 5 — set it high)
exports = client.dubbing.get_export_status(job_id=job_id, limit=100).data.exports
downloads = {(e.target_language, e.export_type): e.download_url for e in exports if e.status == "completed"}
print(downloads)

Quick Start (JavaScript/TypeScript)

import fs from "fs";
import { SarvamAIClient } from "sarvamai";

const client = new SarvamAIClient({ apiSubscriptionKey: process.env.SARVAM_API_KEY });

// 1. Create job — JS wants the RAW WIRE field names src_lang/target_langs
//    here, NOT source_language_code/target_language_codes (see Gotchas —
//    this differs from what Sarvam's own docs sample currently shows)
const created = await client.dubbing.create({
  src_lang: "en-IN",
  target_langs: ["hi-IN", "ta-IN"],
  export_options: ["video", "srt"],
  voice_cloning: true,
  num_speakers: 1,
  job_name: "sample.mp4",
});
const jobId = created.data.job_id;

// 2. Upload to the signed URL (no documented JS upload helper — PUT it yourself)
await fetch(created.data.upload_url, {
  method: "PUT",
  headers: { "Content-Type": "video/mp4", "x-ms-blob-type": "BlockBlob" },
  body: fs.readFileSync("sample.mp4"),
});

// 3. Start — job_id is positional
await client.dubbing.start(jobId);

// 4. Poll live status — also positional
let job;
do {
  job = (await client.dubbing.getLiveStatus(jobId)).data;
  await new Promise((r) => setTimeout(r, 15000));
} while (!["completed", "partial_failure", "failed", "deleted"].includes(job.status));

// 5. Poll export status for download links
const exports = (await client.dubbing.getExportStatus(jobId, { limit: 100 })).data.exports;
console.log(exports.filter((e) => e.status === "completed"));

Gotchas

Full Docs

Fetch voice options, register/tone control, and the full job-lifecycle reference from:

• https://docs.sarvam.ai/llms.txt — comprehensive docs index
• Dubbing Overview
• Job Lifecycle
• Rate Limits