Cloud Cloud Speech-to-Text On-Device

Stay organized with collections Save and categorize content based on your preferences.


Speech-to-Text On-Device enables server-quality speech technology on embedded devices. This feature allows you to run streaming speech recognition fully on-device, without any connection to a network or Google servers. The On-Device solution offers several benefits for this use case when compared to a server-side solution: speech recognition is available even if the device is not connected to the network or network connection is limited, and the user's data does not leave the device.

Key capabilities
High quality transcription Apply Google's advanced deep learning neural network algorithms to automatic speech recognition.
Offline Speech Recognition without Internet connection
Efficient models Deploy efficiently with models that are less than 1 GB in size and consume minimal resources.
Voice Activity Detection Detects start and end of human speech.
Confidence Get confidence estimates on the transcription.
Biasing Guide the recognition using your domain knowledge. See additional documentation

What's Next

  1. Contact Google to get access to On-Device.
  2. See the On-Device quickstart to get started sending a query and inspecting the server's behavior.