Noise reduction in shared workspaces

A system for reducing noise in a shared work environment, the system comprising:

• a processor;
• non-transitory machine-readable media including instructions which, when executed by the processor, cause the processor to:
• receive sets of training data from a plurality of human end-users;
• wherein each set of training data includes a first recording of a recitation of at least one sentence whispered at a low intelligibility and a second recording of a recitation of the at least one sentence spoken normally at a high intelligibility by a human end-user of the plurality of human end-users;
• receive a first audio signal representing a first utterance whispered at a low intelligibility of a first human end-user;
• determine a selected set of training data to be used for speech synthesis, chosen from the received sets of training data, by selecting a set of training data corresponding to the first human end-user using speech recognition of the first audio signal as whispered speech;
• perform automated speech recognition on the first audio signal to produce a first data file identifying at least one word conveyed by the first utterance, based on a recognition model trained using the first recording included in the selected set of training data;
• analyze prosody of the first audio signal representing the received first utterance whispered at a low intelligibility;
• convert the first data file to a second audio signal at a higher intelligibility via implementation of a speech synthesizer trained using the second recording included in the selected set of training data and the analyzed prosody, wherein the second audio signal is more audible than the first audio signal and the second audio signal is synthesized speech representing the at least one word of the first utterance;
• wherein the speech synthesizer causes the second audio signal to have speech characteristics corresponding to those of the first utterance of the first human end-user based on using the selected set of training data to convert the first data file;
• and
• transmit the second audio signal to a second device.

A system for reducing noise in shared work environments, the system comprising:

• a processor;
• non-transitory machine-readable media including instructions which, when executed by the processor, cause the processor to:
• receive sets of training data from a plurality of human end-users;
• wherein each set of training data includes a first recording of a recitation of at least one sentence subvocalized at a low intelligibility and a second recording of a recitation of the at least one sentence spoken normally at a high intelligibility by a human end-user of the plurality of human end-users;
• receive a first audio signal representing a first utterance subvocalized at a low intelligibility of a first human end-user by a wearable device configured to detect subvocalized utterances using electrical signals from at least one of facial muscles and vocal chords of the first human end-user;
• determine a selected set of training data to be used for speech synthesis, selected from the received sets of training data, based on selecting a set of training data corresponding to the first human end-user based on speech recognition of the first audio signal as subvocalized speech;
• perform automated speech recognition on the first audio signal to produce a first data file identifying at least one word conveyed by the first utterance, based on a recognition model trained using the first recording included in the selected set of training data;
• analyze prosody of the first audio signal representing the received first utterance subvocalized at a low intelligibility;
• convert the first data file to a second audio signal at a higher intelligibility via implementation of a speech synthesizer trained using the second recording included in the selected set of training data and the analyzed prosody, wherein the second audio signal is more audible than the first audio signal and the second audio signal is synthesized speech representing a sequence of at least one word included in the first utterance;
• wherein the second audio signal is synthesized to have speech characteristics similar to those of the first utterance by the first human end-user based on using the selected set of training data while using the speech synthesizer to convert the first data file;
• and
• transmit the second audio signal to a second device.