Speech recognition for assisting patients with speech difficulties

A method for impaired speech recognition and/or transcription, comprising:

• training a speech recognition model using a plurality of datasets comprising a plurality of speech samples, thereby generating a fully trained speech recognition model that achieves reductions in word error rate (WER);
• processing, by at least one processor, a user's voice to generate a processed voice file;
• transcribing, by at least one processor executing the fully trained speech recognition model, the processed voice file to generate a transcribed output;
• and
• displaying, via a graphical user interface (GUI), the transcribed output,
• wherein the plurality of speech samples comprise samples from patients with a speech impairment,
• wherein the reductions in WER are achieved by optimizing one or more fine-tuning parameters of the speech recognition model.

A system for impaired speech recognition and/or transcription, comprising:

• a memory storing computer-readable instructions;
• and
• at least one processor to execute the computer-readable instructions to:
• automatically load a virtual agent to join a virtual meeting between a user with a speech impairment and at least one other individual;
• listen, by the virtual agent, to the user's speech when the user communicates to the at least one other individual during the virtual meeting;
• transcribe, by a speech recognition model, the user's speech into a transcribed output that corrects for the speech impairment;
• repeat, by the virtual agent speaking in a synthesized voice, the transcribed output to the at least one other individual;
• and
• track a progression of the speech impediment,
• wherein the speech recognition model is trained using samples from patients with the speech impairment.