English
What customer need does this solve?
Forms are not always convenient. A field technician, factory supervisor, warehouse worker, or manager on the move may know exactly what to say, but typing long explanations or approval notes on a small screen slows the process down.
The customer need is not a standalone voice toy. It is voice inside the workflow fields that already matter: the reason for a request, the note from an approver, and the spoken context that helps a manager decide faster.
What the solution should solve
The solution should support both directions. Employees can speak into a field and have speech-to-text land in the actual textarea. Managers can listen to the request through text-to-speech, dictate their approval notes, and complete the decision on the same record.
The demo proves this on a real Leave Request flow. The employee logs in, opens the leave form, clicks the microphone, dictates the reason, and submits. The manager then opens the approval form for that exact record, listens to the request, dictates an approval comment, and approves it.
Trade-offs and alternatives
The benefit is lower friction for field-heavy or mobile-heavy workflows. The trade-off is that voice features must be tested at the system boundary, not mocked inside the UI. Microphone capture, PCM conversion, transcription, field update, form submit, workflow transition, and database persistence all have to work together.
That testing surfaced a real bug: while recording, the mic button title changed to "Stop recording", making the active button hard to re-locate reliably by title or accessibility tooling. The fix was to add stable test IDs at both call sites, voice-mic-reason and voice-mic-notes. The automated test then used generated Vietnamese WAV samples as fake microphone input in Chromium, exercising the real getUserMedia to /api/transcribe pipeline.
Alternatives include plain typing, mobile dictation outside the app, or a chatbot-style interface. Those can help in narrow cases, but they often detach the spoken content from the workflow record and approval audit trail.
Conclusion
Voice belongs inside operational software when it reduces friction without weakening traceability. In this case, the final database row confirmed the approved status and preserved both the employee reason and manager notes that came from voice input.