An interactive flow from microphone input to live text, a text response, and spoken playback, including Hong Kong Traditional Chinese output for Cantonese.
Open image
This project gives developers and product teams a hands-on way to evaluate voice input and spoken responses for an application. Users can try speech-to-text on its own, or experience speaking, receiving a text reply, and hearing it read aloud. Working with real utterances helps assess whether an interaction feels usable and where speech could replace part of a keyboard-based workflow.
Cantonese output uses Hong Kong Traditional Chinese. Completed segments show both the original result and the adjusted text, allowing names, wording, and phrasing to be checked against what was said. Segment times and counts help with reviewing short dictation, voice-input tests, or meeting transcription assistance one segment at a time. The text still needs to be checked against the speech, particularly with overlapping speakers, background noise, or specialized names.
Product teams can use the same utterance to check recognition, response suitability, and spoken clarity separately, rather than judging only whether any sound was produced.
These flows support evaluating voice integration, checking Cantonese text presentation, and comparing spoken output. Recognition and speech options depend on the configured services. The project provides an interactive voice experience; business use still requires validation with the intended users’ language habits and environment.
I independently designed the voice service, its protocols and system architecture, and built and deployed its frontend and backend. I integrated voice activity detection, segmented recognition, model responses and streaming speech output, with Cantonese text normalization, persisted conversations, project access controls and usage management.
Explore more work with similar capabilities or uses.
Ask about centre services in Cantonese or text, edit spoken questions before sending, follow up and revisit browser-held history in a readable text conversation.
Turn files, web pages, notes and FAQs into a maintainable knowledge assistant, connecting imports, trial questions and conversation review to improve information access for customers and staff.
Connect song practice, singing feedback and teacher-reference comparison with pitch charts, class records and revocable report sharing.
