Skip to content

Assist speech-to-text fails in macOS Companion App while working on iPhone and web #5613

Description

@hepcatguy

Describe the bug

Assist speech-to-text consistently fails in the Home Assistant macOS Companion App with speech-to-text failed.

The same Home Assistant Assist pipeline works correctly on the iPhone Companion App and in Safari when Home Assistant is accessed via HTTPS.

The macOS app does capture and transmit audio. Debug logging shows that microVAD successfully detects both the start and end of speech, but STT returns no transcription and ends with an error.

Environment

  • Home Assistant Core: 2026.9.0
  • Home Assistant macOS Companion App: 2026.9.0 (Build 2026.2874)
  • Assist language: German (de-DE)
  • Tested STT providers: Home Assistant Cloud STT and OpenAI STT
  • Both STT providers fail in the macOS app
  • The same Assist setup works on iPhone
  • Safari via HTTPS also works

Steps to reproduce

  1. Open Assist in the Home Assistant macOS Companion App.
  2. Select the configured Assist pipeline.
  3. Start a voice command.
  4. Speak a short command.
  5. Speech detection starts and stops normally.
  6. Assist returns speech-to-text failed.

Relevant debug output

homeassistant.components.assist_pipeline.audio_enhancer Initialized microVAD
homeassistant.components.assist_pipeline.vad Voice command started
homeassistant.components.assist_pipeline.vad Voice command finished
homeassistant.components.assist_pipeline.pipeline SpeechResult(text=None, result=SpeechResultState.ERROR: 'error')
homeassistant.components.websocket_api.http.connection Received binary message for non-existing handler

The Received binary message for non-existing handler message occurs after the STT result has already returned ERROR, so it may be a consequence of the failed/closed STT stream rather than the initial cause.

Additional information

  • Multiple microphone inputs have been tested on the Mac with the same result.
  • The microphone itself works normally in other macOS applications.
  • Debugging shows that audio reaches Home Assistant and VAD successfully detects speech.
  • The macOS app sends binary audio data during the recording.
  • The failure occurs after speech has ended, when STT should produce the transcription.
  • Changing from Home Assistant Cloud STT to OpenAI STT does not change the behavior.
  • The issue also occurred before enabling the Home Assistant 2026.9 Labs Speech-to-Text processing feature.

This appears to be specific to the macOS Companion App/audio path rather than the configured Assist pipeline or STT provider.

Full logs are intentionally not attached because they contain private Home Assistant Cloud URLs, webhook data, device identifiers, and other sensitive information. Relevant excerpts are included above.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Fields

    Priority

    None yet

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions