#Local Voice Assistant pipeline TTS fails to speak after 2025.5 update

1 messages · Page 1 of 1 (latest)

lethal flint
#

I've been using the "standard" local Voice Assistant set up. Assist Microphone -> openWakeWord -> Whisper (stt.faster_whisper) -> the built-in basic intent system ("Natural Language Processing" conversation.home_assistant) -> Piper (tts.piper).

After upgrading from 2025.4.4 to 2025.5.0 whenever the Voice Assistant doesn't receive a command for a while (more than minute, less than hour), the first command/pipeline always consistently fails to speak the TTS. The audio cue for a response does play, but the TTS speech does not. I haven't found anything of note in the Piper logs or Voice Assistant debug. They all look as if everything worked.

On the second and subsequent commands the TTS speech does play. I'm not sure on the exact timing but after a couple minutes it is still working.

I tried downgrading to 2025.4.4 and this fixes the issue. Then I updated to 2025.5.0 and the issue is back. The issue also persists through 2025.5.1 and 2025.5.2.

EDIT: This is a single device running Home Assistant OS, with a USB microphone/speakerphone (hence Assist Microphone as the "satellite" although it runs on the same device).

lethal flint
#

I can do more debugging but I am a bit lost on where to start.

lethal flint
#

I've narrowed the timing down. After about 9 minutes the TTS still works. After about 10 minutes the TTS fails for the first command/pipeline.

proud rivet
#

There have already been many changes (and more are expected) to the tts system component, as the developers are planning to implement streaming response text generation. There is probably a conflict for your setup. Have you checked with other tts engines?

lethal flint
#

no I haven't, what are the options? Piper is part of the Year of the Voice so I'd assumed it should work

stable sand
#

I have a voice assistant I made myself. I use Cloud and ChatGPT. I have the same problem. It worked perfectly fine for a while, then after an ESPHome update it broke. It doesn't say long answers. But for example, it doesn't always say my morning wake-up routine. That would be a longer text too. the answers are generated. I'll grab a log entry: [17:14:56][W][esp-idf:000][ann_read]: W (1998086) HTTP_CLIENT: Connection timed out before data was ready!
[17:14:56][E][speaker_media_player.pipeline:112]: Media reader encountered an error: ESP_FAIL

uncut rampart
#

hello friends i have the same problem i want to install for my work a local assistant i install all the plugins piper, whisper, speecht, onewakeup, assist mic, but when i go to debug it appears indeterminately as if the wakeup is activated and all the services are frozen i am still a novice but i have a lot of logs with errors.

Traceback (most recent call last):
File "/usr/src/homeassistant/homeassistant/components/assist_satellite/entity.py", line 409, in async_accept_pipeline_from_satellite
await self._pipeline_task
File "/usr/src/homeassistant/homeassistant/components/assist_pipeline/init.py", line 138, in async_pipeline_from_audio_stream
await pipeline_input.validate()
File "/usr/src/homeassistant/homeassistant/components/assist_pipeline/pipeline.py", line 1687, in validate
await asyncio.gather(*prepare_tasks)
File "/usr/src/homeassistant/homeassistant/components/assist_pipeline/pipeline.py", line 877, in prepare_speech_to_text
raise SpeechToTextError(
...<2 lines>...
)
homeassistant.components.assist_pipeline.error.SpeechToTextError: Pipeline error code=stt-provider-missing, message=No speech-to-text provider for: stt.speech_to_phrase

📎 message.txt
lethal flint
#

I'd love to find other users who have the same issue but it seems I'm the only one with the local voice assistant pipeline failing in this specific way? I.e. after 10 minutes of idle the TTS announce fails once but only once, until 10 minutes passes again.