ASR For Disordered Speech
People with speech disorders often have physical disabilities which make it difficult to use normal keyboard-mouse interfaces.
For these people voice-driven interfaces are an attractive alternative even though speech production is limited.
Considered as an ASR problem,
- There will be a small vocabulary of isolated words or short phrases,
- Recognition will be speaker-dependent
- there will be little data to train on (it’s often an effort to speak)
- what data there is will be highly variable
The only way to obtain more data is from a system which is in use
A new methodology for building personal recognisers is needed