feat(widget): add ASR (speech-to-text) support to Floating Widget

[GlobalStar117]

Summary

This PR adds ASR (Automatic Speech Recognition) support to the Floating Widget embedded chat, providing feature parity with the Fullscreen Chat.

Related Issue

Fixes #12709

Changes

• Import AudioButton component into floating-chat-widget.tsx
• Add microphone button to window mode input area
• Add microphone button to full mode input area
• Transcribed text is appended to the input field (allowing users to edit before sending)

Expected Behavior

• Microphone button appears next to the text input in the Floating Widget
• User can click to start/stop recording
• Transcribed text is inserted into the input field
• Graceful handling if microphone permission is denied (handled by existing AudioButton component)

Screenshots

The implementation reuses the existing AudioButton component which already has:

• Voice visualizer during recording
• Recording timer
• Processing indicator
• API call to /sequence2txt endpoint

Testing

1. Embed the Floating Widget on a page
2. Click the microphone button
3. Grant microphone permission
4. Speak and stop recording
5. Verify transcribed text appears in the input field

---

[asiroliu]

Thank you for your submission. Unfortunately, while verifying your changes, we encountered a separate issue that interrupted our testing. We will resume the verification of your PR once this new issue has been resolved.

[Oranggge]

Maybe my two cents will help.

I think this is connected to issue #12602.
I also got this 401 error from this endpoint but in the fullscreen chat.

I managed to solve it locally with the following change:

--- a/api/apps/__init__.py
+++ b/api/apps/__init__.py
@@ -107,7 +107,12 @@ def _load_user():
         try:
             authorization = request.headers.get("Authorization")
             if len(authorization.split()) == 2:
-                objs = APIToken.query(token=authorization.split()[1])
+                token_value = authorization.split()[1]
+                # Try 'token' field first (standard API tokens)
+                objs = APIToken.query(token=token_value)
+                # Fallback to 'beta' field (shared chat tokens)
+                if not objs:
+                    objs = APIToken.query(beta=token_value)
                 if objs:
                     user = UserService.query(id=objs[0].tenant_id, status=StatusEnum.VALID.value)
                     if user:

I didn’t test it extensively, so I might be wrong.