Speech recognition
Transcribe files, streams, calls, meetings, dictation, and application audio inside the selected environment.
Izwi Voice AI Runtime
A private runtime for the core voice AI stack
Run speech recognition, text-to-speech, speaker diarization, and voice workflows close to the data. Evaluate locally with the desktop app and CLI, integrate through familiar /v1 endpoints, and move into supported enterprise deployment when the workflow becomes production infrastructure.
Keep the client. Move the processing boundary.
from openai import OpenAI client = OpenAI( base_url="http://localhost:8080/v1", api_key="not-needed",) transcript = client.audio.transcriptions.create( model="qwen3-asr-0.6b", file=open("audio.wav", "rb"),)Core capabilities
One runtime for private speech workloads
Text-to-speech
Generate voice output without routing every request through an external speech API.
Speaker diarization
Separate speakers in calls, meetings, interviews, and documentation workflows within the same deployment boundary.
Voice workflows
Combine speech services, model selection, and application logic behind a consistent local runtime and API surface.
Product surfaces
Evaluate, integrate, and deploy through the right surface
- Local runtime — run speech models and workflows close to the work.
- OpenAI-compatible API — use familiar /v1 request patterns for available voice endpoints.
- CLI — pull models, start services, inspect configuration, and operate locally.
- Desktop app — a native evaluation and prototyping surface.
- Enterprise packaging — add versioned builds, model bundles, deployment guidance, observability hooks, and support.
Deployment targets
From a developer machine to controlled production infrastructure
We match each deployment to the speech workload, selected models, available hardware, and operating requirements.
- Developer workstation
- Apple Silicon where supported
- NVIDIA server where supported
- Customer VPC
- On-premises
- Edge or offline after feasibility review
- Planned isolated environment
Who it is for
Voice as infrastructure, not a hosted add-on
- software vendors embedding private speech into regulated products
- contact-center and conversation-intelligence platforms
- healthcare, legal, financial, and professional workflow vendors
- local or private agent systems that need a speech layer
- teams replacing a public speech API in production
- edge products that cannot rely on continuous connectivity
Enterprise path
Evaluate openly. Add production support when you need it.
Use the public runtime to validate technical fit. Choose an enterprise deployment when you need versioned releases, automation, security documentation, observability, rollout guidance, or ongoing support.
Need a private language model behind the voice workflow?
Izwi can deploy an open-weight LLM serving layer in the same customer-controlled environment for approved language workloads.