Dictation in ChatGPT web: post-submit processing slower than before, and visualizer state is misleading during load

Two related issues with the dictation feature in the ChatGPT, both surfacing after clicking the checkmark to end recording:

1. Processing time has noticeably increased (regression)
Sometime in the last 1-2 weeks, the wait between hitting the checkmark and seeing the transcribed text appear in the prompt input has roughly 2-3x’d compared to what it used to be. Behavior is consistent across sessions.

2. Audio visualizer continues animating during the post-submit load state
While the audio is being processed (input area in load state), the visualizer keeps doing its left-to-right sweep — the same animation used while the mic is actively recording. The frequency bands also remain at full vertical height. Since the L-to-R sweep is the established “mic is hot” affordance, this reads as if dictation is still capturing audio. I confirmed it isn’t: speaking after the checkmark click is not retroactively streamed into the pre-submit input.

Suggested fixes for #2:

  • Freeze the left-to-right animation in place once the checkmark is clicked. Optionaly, apply a subtle treatment to the frozen state (reduced opacity, or a brightness/desaturation filter) to convey that the captured audio is in a pending/processing state.
  • Only relevant if the left-to-right animation is retained post-submit: invert the frequency bands during the load state — short / vertically-empty bands intuitively read as silence, which matches the actual state (no live capture).