Hey there, thanks for flagging this, after some investigations my hypothesis is that it’s an issue in the cache.
I currently don’t have access to that system, but if it’s what I think it is, will hopefully have it resolved when the Americans wake up.
I know this is a very old thread – but for the last 2 days I am having exactly the problem described here. I tried deleting the cache as mentioned above, and had no change. The audio log shows http 500 errors, such as:
2022-01-27 19:18:09.664 | INFO | 1591 | mycroft.audio.speech:mute_and_speak:127 | Speak: It’s currently clear sky and 20 degrees fahrenheit.
Playing WAVE ‘/tmp/mycroft/cache/tts/Mimic2/a18ba8f65e41175ee80cd18448d887ec.wav’ : Signed 16 bit Little Endian, Rate 22000 Hz, Mono
2022-01-27 19:18:13.293 | INFO | 1591 | mycroft.audio.speech:mute_and_speak:127 | Speak: Today’s forecast is for a high of 20 and a low of -11.
2022-01-27 19:18:13.379 | ERROR | 1591 | mycroft.audio.speech:mute_and_speak:131 | Backend returned HTTP status 500
Playing WAVE ‘/tmp/mycroft/cache/tts/Mimic/1875b6c4b39a73a5bc04b392f045cbae.wav’ : Signed 16 bit Little Endian, Rate 44100 Hz, Mono
It’s actually a little funny sometimes – I asked for the weather in the example above and got the current info in the American voice, and the forecast in the British one. 
Just wondering is anybody has any insights as to what I can do to stop this. I don’t really want to switch to the google voice … kind of defeats the purpose of choosing Mycroft…
I seem to be having the same problem with the voice switching intermittently between the two voices. Also seeing 500 errors. Recently switched to Raspberry Pi OS 64bit as a test so I thought it was something with my setup. Didn’t happen with the original Picroft image.
When I mentioned that to @ChrisVeilleux last Friday is answered this (on the chat).
Maybe the issue is back again 
Thanks for the info – it was still happening as of yesterday (2/1/22). I ended up setting up coqui on an old laptop last night. It’s a little slower, but at least it gives a consistent and understandable voice. I’ll keep using that.
This has been a long-running problem and needs to be handled with a less-fragile solution.
Totally agree, monitoring is the key!
