> For clean Markdown of any page, append `.md` to the page URL. > For a complete documentation index, see https://docs.sarvam.ai/llms.txt. > For full documentation content in one file, see https://docs.sarvam.ai/llms-full.txt. > For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.sarvam.ai/_mcp/server. # How to enable speaker diarization > Identify and distinguish between multiple speakers in audio using the Batch API. **Batch API only:** Speaker diarization is only available through the Batch API, not the REST or Streaming APIs. Speaker diarization identifies and labels different speakers in your audio, making it easy to know "who said what." This is ideal for meetings, interviews, podcasts, and call center recordings. ### Key Features * Automatic speaker detection * Support for up to 10 speakers * Speaker-wise transcription with timestamps ### Parameters | Parameter | Type | Description | | ------------------ | ------- | --------------------------------------------- | | `with_diarization` | boolean | Enable speaker diarization (default: `false`) | | `num_speakers` | integer | Expected number of speakers (optional, 1-10) | If you don't specify `num_speakers`, the model will automatically detect the number of speakers. ### Example Code ```python from sarvamai import SarvamAI client = SarvamAI(api_subscription_key="YOUR_SARVAM_API_KEY") # Create batch job with diarization job = client.speech_to_text_job.create_job( model="saaras:v3", language_code="hi-IN", mode="transcribe", with_diarization=True ) # Upload audio files job.upload_files(file_paths=["meeting_recording.mp3"]) # Start processing job.start() # Wait for completion job.wait_until_complete() # Download results job.download_outputs(output_dir="./output") ``` ```javascript import { SarvamAIClient } from "sarvamai"; const client = new SarvamAIClient({ apiSubscriptionKey: "YOUR_SARVAM_API_KEY" }); // Create batch job with diarization const job = await client.speechToTextJob.createJob({ model: "saaras:v3", languageCode: "hi-IN", mode: "transcribe", withDiarization: true }); // Upload audio files await job.uploadFiles(["meeting_recording.mp3"]); // Start processing await job.start(); // Wait for completion await job.waitUntilComplete(); // Download results await job.downloadOutputs("./output"); ``` ```bash # Step 1: Create the job (parameters go inside "job_parameters") curl -X POST https://api.sarvam.ai/speech-to-text/job/v1 \ -H "api-subscription-key: " \ -H "Content-Type: application/json" \ -d '{ "job_parameters": { "model": "saaras:v3", "language_code": "hi-IN", "mode": "transcribe", "with_diarization": true } }' # Response: {"job_id": "", ...} # Step 2: Get a presigned upload URL for each audio file curl -X POST https://api.sarvam.ai/speech-to-text/job/v1/upload-files \ -H "api-subscription-key: " \ -H "Content-Type: application/json" \ -d '{"job_id": "", "files": ["meeting_recording.mp3"]}' # Response: {"upload_urls": {"meeting_recording.mp3": {"file_url": ""}}} # Step 3: Upload the audio file to the presigned URL # (Content-Type must be an audio MIME type matching your file, e.g. audio/wav for .wav) curl -X PUT "" \ -H "x-ms-blob-type: BlockBlob" \ -H "Content-Type: audio/mpeg" \ --data-binary @meeting_recording.mp3 # Step 4: Start the job curl -X POST https://api.sarvam.ai/speech-to-text/job/v1//start \ -H "api-subscription-key: " # Step 5: Poll status until "job_state" is "Completed" curl https://api.sarvam.ai/speech-to-text/job/v1//status \ -H "api-subscription-key: " # Step 6: Get presigned download URLs for the output files # (output file names come from job_details[].outputs[].file_name in the status response) curl -X POST https://api.sarvam.ai/speech-to-text/job/v1/download-files \ -H "api-subscription-key: " \ -H "Content-Type: application/json" \ -d '{"job_id": "", "files": ["0.json"]}' ``` ```python from sarvamai import SarvamAI client = SarvamAI(api_subscription_key="YOUR_SARVAM_API_KEY") # Create batch job with known speaker count job = client.speech_to_text_job.create_job( model="saaras:v3", language_code="en-IN", mode="transcribe", with_diarization=True, num_speakers=3 # Interview with 3 participants ) job.upload_files(file_paths=["interview.mp3"]) job.start() job.wait_until_complete() job.download_outputs(output_dir="./output") ``` ```javascript import { SarvamAIClient } from "sarvamai"; const client = new SarvamAIClient({ apiSubscriptionKey: "YOUR_SARVAM_API_KEY" }); // Create batch job with known speaker count const job = await client.speechToTextJob.createJob({ model: "saaras:v3", languageCode: "en-IN", mode: "transcribe", withDiarization: true, numSpeakers: 3 // Interview with 3 participants }); await job.uploadFiles(["interview.mp3"]); await job.start(); await job.waitUntilComplete(); await job.downloadOutputs("./output"); ``` ```bash # Create the job with a known speaker count, then follow the same # upload → start → poll → download flow shown in the Basic Diarization tab. curl -X POST https://api.sarvam.ai/speech-to-text/job/v1 \ -H "api-subscription-key: " \ -H "Content-Type: application/json" \ -d '{ "job_parameters": { "model": "saaras:v3", "language_code": "en-IN", "mode": "transcribe", "with_diarization": true, "num_speakers": 3 } }' ``` ### Output Format When `with_diarization=True` is passed, the response includes a `diarized_transcript` field with speaker information: ```json { "request_id": "20260130_d8d2c0e6-1eb6-4982-8045-b267d5165c44", "transcript": "Full transcript text...", "timestamps": { "words": ["Hello, how can I help you today?", "I have a question about my order."], "start_time_seconds": [0.01, 2.8], "end_time_seconds": [2.5, 5.2] }, "diarized_transcript": { "entries": [ { "transcript": "Hello, how can I help you today?", "start_time_seconds": 0.01, "end_time_seconds": 2.5, "speaker_id": "0" }, { "transcript": "I have a question about my order.", "start_time_seconds": 2.8, "end_time_seconds": 5.2, "speaker_id": "1" } ] }, "language_code": "en-IN" } ``` Each entry contains: * `transcript`: The text spoken by the speaker * `start_time_seconds`: When the speaker started speaking (float) * `end_time_seconds`: When the speaker stopped speaking (float) * `speaker_id`: Unique identifier for the speaker (e.g., "0", "1") ### Use Cases | Use Case | Recommended Settings | | ---------------------- | --------------------- | | Call center recordings | `num_speakers=2` | | Meetings | Let model auto-detect | | Interviews | Specify exact count | | Podcasts | `num_speakers=2-4` | Speaker diarization is available via the Batch API and has separate pricing. See the [pricing page](/api/pricing) for detailed rates. → [Full Batch API Documentation](/api/api-guides-tutorials/speech-to-text/batch-api)