Transcribe audio with speaker labels - AudioPod AI

What you get: a transcript split into timed segments, each tagged with who spoke it (SPEAKER_00, SPEAKER_01, …). The language is detected automatically, or you can set it. The same job also exports SRT, VTT, TXT, DOCX and PDF. Input: a public URL of an audio or video file, or a local file.

1. Install and set your key

pip install audiopod
export AUDIOPOD_API_KEY="ap_..."

npm install audiopod
export AUDIOPOD_API_KEY="ap_..."

Get a key in your dashboard. New accounts start with free credits.

2. Transcribe

from audiopod import Client

client = Client()  # reads AUDIOPOD_API_KEY

job = client.transcription.transcribe(
    url="https://example.com/interview.mp3",  # or audio_file="interview.mp3"
    speaker_diarization=True,
    wait_for_completion=True,
)

transcript = client.transcription.get_transcript(job["id"])  # JSON
for seg in transcript["segments"]:
    print(f'{seg["start"]:7.1f}s  {seg.get("speaker_label", "")}: {seg["text"].strip()}')

import AudioPod from "audiopod";

const client = new AudioPod(); // reads AUDIOPOD_API_KEY

// Starts the job and waits until it completes
const job = await client.transcription.transcribe({
  url: "https://example.com/interview.mp3",
  speakerDiarization: true,
});

// Subtitles straight from the finished job: "srt", "vtt", "txt" or "json"
const srt = await client.transcription.getTranscript(job.id, "srt");
console.log(srt);

For a local file in Node, use client.transcription.upload({ file: "./interview.mp3", speakerDiarization: true }), then client.transcription.waitForCompletion(job.id).

What comes back

Field	Meaning
`job["id"]` / `job.id`	The transcription job id. Use it to fetch the transcript in any format.
`status`	`COMPLETED` once the transcript is ready (`FAILED` raises an error in both SDKs).
`segments[]` (JSON transcript)	`start`, `end` (seconds), `text` and `speaker_label` for each segment.
`detected_language`	The language AudioPod detected, when you didn’t set one.

Tips

Set language="en" (Python) or language: "en" (Node) when you know it; detection is automatic otherwise.
In Node, minSpeakers and maxSpeakers bound the speaker count when you know roughly how many people talk.
Want to be told when a long recording finishes instead of waiting? Register a webhook.
Live audio from a microphone or call: see live transcription.

From an AI agent

Connected to the AudioPod MCP server? Ask: “Transcribe https://example.com/interview.mp3 with speaker labels as SRT.” The agent calls transcribe_audio with diarize: true and format: "srt", then check_job_status until it is done. Full endpoint reference: Speech-to-Text.