Speech-to-phrase training fails with aiohttp.client_exceptions.WSMessageTypeError

I’m trying to install the new add-on speech-to-text. When is start the training it fails with:

  File "/usr/src/.venv/lib/python3.11/site-packages/aiohttp/client_ws.py", line 386, in receive_str
    raise WSMessageTypeError(
aiohttp.client_exceptions.WSMessageTypeError: Received message 8:1000 is not WSMsgType.TEXT

Update: log entries from the supervisor:

2025-02-15 08:47:55.844 INFO (MainThread) [supervisor.api.proxy] Home Assistant WebSocket API request initialize
2025-02-15 08:47:55.845 INFO (MainThread) [supervisor.api.proxy] WebSocket access from core_speech-to-phrase
2025-02-15 08:47:55.847 INFO (MainThread) [supervisor.api.proxy] Home Assistant WebSocket API request running
2025-02-15 08:47:57.984 INFO (MainThread) [supervisor.api.proxy] Home Assistant WebSocket API error: Cannot proxy websocket message of unsupported type: 258
2025-02-15 08:47:57.985 INFO (MainThread) [supervisor.api.proxy] Home Assistant WebSocket API for core_speech-to-phrase closed
2025-02-15 08:47:59.611 INFO (MainThread) [supervisor.api.proxy] Home Assistant WebSocket API request initialize
2025-02-15 08:47:59.612 INFO (MainThread) [supervisor.api.proxy] WebSocket access from core_speech-to-phrase

Same experience here, as soon as install completes and add-in starts, a ton of errors like below

aiohttp.client_exceptions.WSMessageTypeError: Received message 258:WebSocketError(<WSCloseCode.MESSAGE_TOO_BIG: 1009>, ‘Message size 11866656 exceeds limit 4194304’) is not WSMsgType.TEXT

[13:26:40] INFO: Service exited with code 1 (by signal 0)

I created a bug report here: Training fails with error Received message 8:1000 is not WSMsgType.TEXT · Issue #14 · OHF-Voice/speech-to-phrase

I had seen the post by OP, but thought, that is was a different problem (‘MESSAGE_TOO_BIG’ vs ‘8:1000’), and had posted OHF-Voice/speech-to-phrase container error "MESSAGE_TOO_BIG" on start .

If this is the same problem, can the mods merge the threads somehow?

It might be the same issue but I don’t know. So just leave it as it is.

This is also discussed on github: WSCloseCode.MESSAGE_TOO_BIG during training (first startup) · Issue #6 · OHF-Voice/speech-to-phrase

This is fixed with HA 2025.3.0.