Use cases · Podcasts
Podcast transcription API with speaker labels
Speech Revolutions transcribes podcast episodes with every turn labelled by speaker and every word timestamped, so transcripts, show notes, clips and captions come from one API call. Diarization is included in the per-minute price and is where we measure best against other APIs.
from speechrevolutions import SpeechRevolutions
client = SpeechRevolutions()
result = client.transcribe(
"episode-142.mp3",
custom_vocabulary=["Zephyr", "diarization"],
)
NAMES = {"SPEAKER_1": "Host", "SPEAKER_2": "Guest"}
for u in result.utterances:
print(f"{NAMES.get(u.speaker, u.speaker)}: {u.text}")Where it goes wrong
A podcast transcript is read as a conversation. When a guest's answer is credited to the host, or a speaker change is missed mid-sentence, the transcript says something nobody said, and every quote, clip or social post pulled from it carries the wrong name.
Episodes are long, often an hour or more, with two to four voices that talk over each other, laugh and interrupt. That is exactly where speaker labelling breaks first.
What the API gives you
- Speaker turns
- result.utterances gives one entry per turn with speaker, text, start and end: render it straight into a transcript page.
- Word timestamps
- Every word has start and end times, so you can cut audiograms and clips on word boundaries and sync a highlighted transcript to the player.
- Captions
- Ask for output_type="srt" or "vtt" and the server returns the caption file.
- Long episodes
- Files up to 10 GB through the SDK, with live progress while they process.
- Show vocabulary
- custom_vocabulary takes your show's names and jargon so they are spelled the way you spell them.
Measured
On VoxConverse, talk shows and interviews with two to four speakers, Speech Revolutions measured 3.6% diarization error rate, against 9.8% for AssemblyAI and 10.6% for Deepgram.
Full results and method: speaker diarization measured across five APIs and all benchmarks.
Pricing
$0.003 per audio minute, $0.18 an hour, billed by the second. Speaker labels, word timestamps and every output format are included. New accounts get $10 of credit with no card required. Pricing details.