Multi-Speaker Transcription

Convert multi-speaker audio into transcription results with speaker information — ideal for meetings, interviews, and podcasts.

1. Create an API Key

   Open the API Key management platform, create an API Key, and save it as an environment variable. Do not write the API Key into client-side code or commit it to a repository.

   MOSS_API_KEY=<your_api_key_here>

2. Prepare the request input

   Prepare a local multi-speaker audio file and submit the file and transcription parameters together using multipart/form-data; pass diarize=true to enable speaker recognition.

   FieldTypeRequirementDescription

   model: stringRequired

   Currently, use moss-transcribe-diarize-pro.

   file: fileRequired

   A local audio file uploaded using multipart/form-data.

   diarize: booleanRequired

   Set to true to enable speaker recognition.

   response_format: stringOptional

   Transcription result format. Available values are json and text; the default is json. Use json to return speaker and segments.

   model=moss-transcribe-diarize-pro
   file=@./audio.mp3
   diarize=true
   response_format=json

3. Send your first request

   Multi-speaker transcription

   curl -X POST "https://api.mosi.cn/v1/audio/transcriptions" \
     -H "Authorization: Bearer $MOSS_API_KEY" \
     -F "model=moss-transcribe-diarize-pro" \
     -F "file=@./audio.mp3" \
     -F "diarize=true" \
     -F "response_format=json"

   import os
   import requests
   
   with open("./audio.mp3", "rb") as audio_file:
       resp = requests.post(
           "https://api.mosi.cn/v1/audio/transcriptions",
           headers={"Authorization": f"Bearer {os.environ['MOSS_API_KEY']}"},
           data={
               "model": "moss-transcribe-diarize-pro",
               "diarize": "true",
               "response_format": "json",
           },
           files={"file": audio_file},
       )
   
   print(resp.status_code)
   print(resp.json())

   import { readFile } from 'node:fs/promises'
   
   const form = new FormData()
   form.append('model', 'moss-transcribe-diarize-pro')
   const audio = new Uint8Array(await readFile('./audio.mp3'))
   form.append('file', new Blob([audio], { type: 'audio/mpeg' }), 'audio.mp3')
   form.append('diarize', 'true')
   form.append('response_format', 'json')
   
   const response = await fetch('https://api.mosi.cn/v1/audio/transcriptions', {
     method: 'POST',
     headers: { Authorization: `Bearer ${process.env.MOSS_API_KEY}` },
     body: form,
   })
   
   console.log(response.status)
   console.log(await response.json())

   Streaming output

   Add stream=true to the same multipart request to read transcription events continuously via SSE. A streaming request cannot also pass async=true.

   curl -N -X POST "https://api.mosi.cn/v1/audio/transcriptions" \
     -H "Authorization: Bearer $MOSS_API_KEY" \
     -H "Accept: text/event-stream" \
     -F "model=moss-transcribe-diarize-pro" \
     -F "file=@./audio.mp3" \
     -F "diarize=true" \
     -F "stream=true" \
     -F "response_format=verbose_json"

4. Retrieve the task result

   For synchronous calls, read results such as text and segments from the response; when using SSE, read through the transcript.text.done event. Passing async=true returns a task_id; query the task result via GET /v1/audio/tasks/{task_id}.

   curl https://api.mosi.cn/v1/audio/tasks/$TASK_ID \
     -H "Authorization: Bearer $MOSS_API_KEY"

5. Verify success

   Success indicator: you obtain a transcription result with speaker information. The success response may include segments, where each segment contains start, end, text, and speaker.

Next steps