Managed Cloud
Run through OrbitalsAI's managed production infrastructure.
The Infrastructure Behind African Voice AI
High-accuracy transcription, translation, and speech generation across African and global languages, with broad accent support and low latency via API and SDKs, all powered by Perigee-1.
$15 free credit · No credit card required
MINI PLAYGROUND
See Perigee-1 work.
Start with your voice or a few words.
Experience Perigee-1, right here.
Input
Tap to record.
Transcript
Ready when you are.
- Speech recognition languages
- 10
- Voice & text languages
- 19
- Translation directions
- 342
- Realtime end-to-end latency
- ~200 ms
One family across the language stack.
Use each model independently or compose them into complete multilingual speech and language workflows.
Recorded audio
To sai a gyara zama, don sauraron cikakkun labarai…
10 languages. Recorded transcription, SRT output and support for real-world audio conditions.
Explore Speech Transcription →Why OrbitalsAI
Language infrastructure should adapt to people, not the other way around.
Real conversations do not stay neatly inside one language. People switch between languages, carry regional accents and speak in conditions that rarely resemble clean benchmark audio.
Perigee-1 is built around that reality.
Natural code-switching
Okay, zan turo maka details ɗin tomorrow.
From voice to understanding — and back again.
Stage 01
Audio
input
NaijaVoices benchmark
Built on evidence, not claims.
Speech recognition performance varies by language, dataset and recording condition. We publish results so developers can evaluate Perigee-1 language by language.
Perigee-1 does not win every benchmark. That is why we evaluate language by language rather than reduce performance to one headline number.
See the full Perigee-1 evaluation →Lower WER is better
Perigee-112.60
Sahara16.73
Gemini 3.0 Flash29.13
MMS-1B39.37
Whisper Large-v396.99
GPT-4o Audio119.74
Lower WER is better
For developers
From API key to first result in minutes.
Use the Python SDK for recorded speech. REST APIs and WebSockets cover the wider Perigee-1 family.
REST API · WebSockets · Python SDK
quickstart.py · Python
import os
import orbitalsai
client = orbitalsai.Client(
api_key=os.environ["ORBITALSAI_API_KEY"]
)
transcript = client.transcribe(
file_path="call.mp3",
language="hausa",
model_name="Perigee-1-transcribe",
)
print(transcript.text)Production performance
Built beyond the demo.
~200 ms
Realtime end-to-end latency
- Batch transcription
- Benchmark: a 30-minute recording processed in under 30 seconds.
- Text-to-speech
- Benchmark: under 1050 ms time to first audio.
- Generation
- Up to 5× faster than realtime.
- TTS throughput
- 82.4 characters per second.
Built for workflows like these.
Examples of how the Perigee-1 family can be composed. These are not customer deployments.
Customer supportUnderstand conversations while they happen.
- Call
- Realtime transcription
- Translation
- Redaction
- Summary
- CRM
Financial servicesTurn multilingual conversations into structured workflows.
- Conversation
- Transcript
- PII redaction
- Structured extraction
- Internal systems
MediaMove from audio to multilingual publishing.
- Audio
- Transcript
- SRT
- Translation
- Localized voice
Public servicesMake information accessible across languages.
- Information
- Translation
- Speech
- Multilingual access
Deploy it where your infrastructure requires.
Customer VPC
Deploy inside your cloud environment for greater network and data control.
On-premise
Run within customer-controlled infrastructure for workloads that cannot leave the environment.
Security and compliance
Status is shown as it stands today.
- NDPA
- Available
- SOC 2
- In progress
- ISO 27001
- In progress
Start building without talking to sales.
No credit card required. Create an account, generate an API key and start testing Perigee-1.
From OrbitalsAI
Research, product and engineering notes.
Build language products for more of the world.
Start with $15 in free credits. No credit card required.