PreviewLive API on commercial engines; proprietary acoustic models remain research in development.Learn more →

Platform Preview

AMBIE combines a live commercial-engine API with proprietary acoustic-intelligence research.

AMBIE is a platform for acoustic intelligence — the goal is 90-95% transcription accuracy in environments where traditional ASR fails (ICUs, manufacturing floors, busy restaurants, court reporting). That is a goal, not a measured result. The proprietary models meant to investigate it remain research in development, with no measured result or committed release date. Read the research agenda.

In the meantime, the API itself is live. The endpoint shapes, auth, retry behavior, webhook signatures, SDKs, billing, and developer experience are stable enough to integrate today. Behind the scenes the inference is currently served by commodity off-the-shelf models. We’re honest about that.

What’s under the hood today

EndpointTodayResearch direction (unvalidated)
TranscribeDeepgram Nova-3 / OpenAI WhisperProprietary acoustic recognition under research
TranslateLlama 3.1 70BNo swap planned
TTS (Aura 2)Deepgram Aura 2Proprietary synthesis under consideration
TTS (MeloTTS)MeloTTS via Cloudflare AINo swap planned
EmbeddingsBGE familyNo swap planned
RerankBGE RerankerNo swap planned
SentimentDistilBERT-SST-2No swap planned
SummarizeLlama 3.3 70BNo swap planned
ModerateLlama Guard 3No swap planned
Detect languagesmall classifierNo swap planned

The transcription and TTS endpoints are the intended focus of proprietary research. Translation, embeddings, rerank, sentiment, summarize, moderate, and detect-language are utility endpoints we offer because they’re commonly chained with audio workflows; we don’t plan proprietary versions of those.

Why integrate now?

What would a proprietary release require?

A planned proprietary release would add an engine: "ambie" option (subject to validation and release documentation). Today’s deepgram and whisper options are intended to remain available for cost-sensitive workloads.

Pricing for the AMBIE engine will be announced before GA and will reflect the proprietary R&D and serving costs of a model built for noisy environments. Today’s rates are passthrough-COGS-based and apply only to the preview engines (deepgram, whisper); plan accordingly when budgeting for the AMBIE engine swap.

Email cisco@ambie.ai if you’re an early-access candidate (high audio volume, noisy-environment use case, willingness to share feedback during beta).

AMBIE is a product of RoamingPigs, Inc. (Las Vegas, NV).