Recognize spoken input.
Speech-to-text supplies the transcript. Your application interprets the request and decides what work follows.
Let your application listen and speak.
Turn a spoken request into text and your application's reply into audio.
Recognition and synthesis run on Origon's speech engine and GPU infrastructure.
Your application leads the conversation.
Generate a spoken response.
Text-to-speech reads the response your application produces. Combine the two for a spoken exchange.
Speech in Origon’s native voice stack
Respond when people interrupt.
Sub-300ms
Voice round trip on Origon's native voice stack.
In Origon's native voice stack, an agent can pause and respond when a person interrupts. Speech works with the media infrastructure to support that exchange.
This figure describes the native voice stack on Origon infrastructure. It is not a standalone STT or TTS API latency guarantee. Evaluate latency for your application and configuration.
Capacity planning
Handle the conversations you expect.
The number of simultaneous recognition and synthesis tasks depends on the GPU hardware allocated to the workload.
Origon runs the reserved Speech service; your team runs the application. Capacity and operating terms follow the agreed workload.
Voice Network
Bring voice to phone calls.
Add Voice Network when the application needs phone numbers, PSTN or SIP connectivity.
Speech and Voice Network are separate services. Use either on its own or combine them for a voice application.
Evaluation
Test recognition and speech with your audio.
Confirm the languages, voice options and interfaces for your configuration. Use representative terminology and audio conditions to assess recognition and generated speech.
For a complete voice application, include response time and interruption handling. The supported configuration, capacity and delivery timing are agreed before reservation.
Language and terminology
Audio conditions
Interaction behavior
Measurement boundary