What is Ultrafield
Ultrafield is Voice AI infrastructure: one API for speech recognition, speaker diarization and identification, audio understanding, and LLM processing.
Ultrafield turns audio into accurate transcripts and structured understanding through a composable API platform.
Use it for automatic speech recognition (ASR), speaker diarization and identification, transcript formatting, entity detection, sentiment analysis, and language-model processing. Route work to Ultrafield's proprietary models or supported open-weight models without rebuilding the rest of your application.
Early access documentation
Ultrafield is taking shape now. These docs describe the intended product surface and clearly mark upcoming capabilities. Exact endpoint paths, schemas, limits, model availability, and public rates are supplied with API access as those contracts stabilize.
One path from audio to understanding
An Ultrafield pipeline has three layers:
| Layer | What it does |
|---|---|
| Speech | Transcribes files, URLs, or real-time audio with the selected ASR model |
| Speakers and understanding | Adds speaker turns, known-speaker identity, formatting, entities, sentiment, and other processors |
| Language models | Transforms speech-derived data into summaries, extraction, classification, or application-specific output |
Choose only the stages a request needs. A captioning product can stop after streaming ASR and formatting; a conversation-intelligence product can continue through diarization, entities, sentiment, and an LLM.
Models
Ultrafield is developing proprietary speech models and provides access to supported open-weight models including Whisper large-v3, Whisper large-v3-turbo, and pyannote-based speaker processing.
Hosted GPT-OSS-120B is available for downstream language processing. A gateway for third-party hosted models from providers such as OpenAI and Anthropic is upcoming.
Designed for production
The platform is built around four constraints:
- Fast inference for responsive products
- Accuracy on the audio that matters to your application
- Cost-effective processing at sustained volume
- Explicit, observable composition instead of hidden processing bundles
Start here
Quickstart
Understand the request lifecycle from audio input to structured output.
Models
Choose between proprietary and open-weight model routes.
Pipelines
Compose transcription, speaker processing, understanding, and LLM stages.
Get API access
Share your audio profile, volume, latency target, and desired output.