About Speech Revolutions
Speech Revolutions LLC builds a batch speech-to-text API. You send audio, we return a transcript with word-level timestamps and speaker labels, in JSON, TXT, SRT, VTT, DOCX or PDF. That is the whole product.
What we build
A transcription API for teams processing recorded audio at volume — meeting and call recordings, media archives, podcasts, research interviews. It is batch, not streaming: files go in, transcripts come out, and the work is done asynchronously so it can be scheduled onto whatever compute is cheapest at the time.
Speaker diarization is the part we are best at. Our measured diarization error rate is less than half that of AssemblyAI, Deepgram and ElevenLabs on the same public datasets, scored with the collar stated rather than omitted — the full method and results are published, because a benchmark you cannot reproduce is marketing.
Automatic language detection covers 99+ languages, including mid-file switching. Diarization, timestamps and every output format are included in the per-minute price rather than sold as add-ons.
How it runs
The GPU fleet is deliberately multi-cloud — AWS, OVHcloud, Hyperstack, Novita and SaladCloud — and jobs are scheduled across it by price and availability. That is the reason the rate is $0.003 per audio minute rather than several times that: we are not paying one vendor's list price for every second of inference.
Turnaround is a median of 48 seconds per hour of audio, measured end to end across 20,261 production jobs over the seven days to 2026-09-17. We publish that figure instead of a speed multiple because a multiple taken from one favourable file is not a number anyone can plan against.
Audio and transcripts are deleted within 30 minutes of a job finishing. We are not a training-data business; your audio is not a second product. The subprocessor list names every service that can receive customer data.
Scale today
- 5,000+
- Jobs per day
- 2,500+
- Audio hours per day
- 1M+
- Minutes processed
Who is behind it
Speech Revolutions was founded by Ameen Ahmed, who designed the diarization pipeline and the multi-cloud scheduler the service runs on, and writes most of what appears on the engineering blog. The company is a limited liability company registered in the United States, founded in February 2026.
It is a small team, which is worth saying plainly rather than implying otherwise: you are talking to the people who wrote the thing. Contact sales or support reaches them directly.
Where we are going
Real-time streaming, an economy tier at $0.0015 per minute, EU data residency and SOC 2 are all on the public roadmap, with what is live today separated from what is not. We would rather publish the list with dates we can be held to than describe everything as available now.