Autonomous Speech Recognition
The October 2021 Autonomous Speech Recognition release introduced self-supervised models trained on 1.1 million hours of audio, up from 30,000 hours in the earlier approach.
Access not confirmed
Broader audio before supervised recognition
The launch studied recognition across African American voices and children’s speech. Its aim was broader representation in training rather than dependence only on manually transcribed recordings. The published comparisons describe specific evaluation datasets, not freedom from all recognition bias.Speechmatics: Autonomous Speech Recognition launch
Practical guidance. This release explains the move toward Speechmatics’ later self-supervised systems. It is a historical model generation, not another current API selector.Speechmatics: Autonomous Speech Recognition launch
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Access not confirmed |
| Launch date | |
| Model identifiers | Autonomous ASR |
| Training source described at launch | Unlabelled internet audio, including podcasts and social mediaSpeechmatics: Autonomous Speech Recognition launch |
| Access scope | Historical software generation; an exact current checkpoint is not identifiedSpeechmatics: Autonomous Speech Recognition launch |
Related models
- Ursa 2
Speechmatics · Ursa
Historical recognition generation that expanded self-supervised training and language coverage.
Launch date Access not confirmed - Ursa
Speechmatics · Ursa
Historical recognition generation with self-supervised audio representations and separate decoding components.
Launch date Access not confirmed