If you are building AI applications today, you've probably used OpenAI's Whisper API. It’s incredibly accurate and easy to use. But as my audio processing volume grew, I hit two massive pain points:
- It's expensive at scale ($0.006 per minute).
- No native speaker diarization (It just returns a giant wall of text, without knowing who is speaking).
To solve this for my own projects, I ended up building an infrastructure that addresses both issues. Today, I'm making it public: FreeAudioToText API.
What it does
It is a drop-in replacement for the official OpenAI SDK. It processes audio with Whisper-level accuracy but adds highly accurate Speaker Diarization (SPK_1, SPK_2) out of the box, especially for English and Chinese.
Oh, and it costs $0.003 per minute — exactly 50% cheaper than OpenAI.
Zero Code Changes Required
I know developers hate learning new SDKs. So I made sure the API is 100% compatible with the official openai-python package.
You just need to change the base_url and your api_key:
from openai import OpenAI
# 1. Just change the base_url and api_key!
client = OpenAI(
api_key="fat-xxxxxxxxx",
base_url="https://api.freeaudiototext.com/v1"
)
# 2. Use the exact same code you already wrote
transcript = client.audio.transcriptions.create(
model="whisper-1",
file=open("meeting.mp3", "rb"),
response_format="diarized_json" # We support speakers!
)
print(transcript.segments[0].speaker) # Output: "SPK_1"
Try it out
I also just published the official wrapper on PyPI if you want to make sure your dependencies are aligned: pip install audiototext-optimizer. (Check it out on PyPI)
If you're building transcription apps, meeting summarizers, or podcasts tools, give it a try. There are no monthly fees, it's strictly pay-as-you-go starting with a few bucks for a Hobby tier.
👉 Documentation & API Keys: freeaudiototext.com/speech-to-text-api
I'd love to hear your feedback in the comments! What features should I add next?
Top comments (0)