On a phone in one hand, typing a long question into a chat box is friction. Visitors abandon at the first thumb fumble. On a desktop with hands full, the same problem shows up. Voice input is the obvious answer, but most WordPress chatbots either skip it or rely on a third-party SaaS that adds latency and privacy concerns.

SleekAI uses the Web Speech API built into modern browsers for fast in-page transcription. The transcript appears in the input field as the visitor talks, so they can edit before sending. For sites that need higher accuracy or wider language coverage, SleekAI ships an optional Whisper-based fallback that runs against your own provider key, with audio uploaded to wp-content/uploads and deleted after transcription.

The hard parts are background noise, accents, and language detection. The Web Speech API is fast but variable. Whisper is more accurate but slower and tokenized. SleekAI lets you pick the default per chatbot, and visitors can switch on the fly. Generic chatbots that bolt on voice usually pick one path with no fallback, and the failure mode is a silent transcript box where nothing arrives.

Voice input is the difference between a chatbot anyone can use and a chatbot only patient typists can use. On mobile the keyboard is the bottleneck. With voice, asking a complete question takes the same time as saying it.

That shaves seconds off every conversation, and seconds determine whether a visitor starts a chat or gives up. Accessibility is the other half of the case. Visitors with motor or vision differences often cannot type a fluent paragraph into a small input field.

Voice input is not a nice-to-have for them, it is the only path. By using the browser's native Web Speech API as the default, SleekAI gives them a first-class experience without any extra assistive software. The Whisper fallback exists because the Web Speech engines, while fast, sometimes miss heavy accents, low-volume speech, or noisy backgrounds.

When accuracy matters more than latency, the same chatbot can route audio to a higher-quality model through your existing provider key. That tradeoff is yours to tune per chatbot, not a vendor decision. Operationally voice also brings a different shape of question.

People say things out loud that they would never type. The transcripts often read more naturally, which gives the model better signal to work with, and gives you better signal about what your audience actually cares about.