14:02:31.289 - __main__:handle_record_begin:36 - INFO - Begin Recording...
14:02:32.524 - mycroft.client.speech.listener:transcribe:231 - ERROR - list index out of range
14:02:32.524 - mycroft.client.speech.listener:transcribe:232 - ERROR - Speech Recognition could not understand audio
The frequency of occurrence is often.
hippyk
All 18 comments
Is it being intentionally triggered or is it false positives?
el-tocino
on 24 Jul 2019
Is it being intentionally triggered or is it false positives?
I'm sorry I don't quite understand what you mean. I'm just doing the regular operation.
By the way, I don't have a good internet connection in China.
hippyk
on 24 Jul 2019
Are you saying "Hey mycroft, do something" and it's responding with the log message above? Or are you not saying anything and it's randomly activating?
el-tocino
on 24 Jul 2019
Are you saying "Hey mycroft, do something" and it's responding with the log message above? Or are you not saying anything and it's randomly activating?
The first case. If I wake it up and say nothing, it's almost certain to happen.
hippyk
on 24 Jul 2019
Are you using a custom wake word or just "hey mycroft"?
el-tocino
on 24 Jul 2019
Are you using a custom wake word or just "hey mycroft"?
I send a wake-up signal directly by "touch startListening"
hippyk
on 24 Jul 2019
No idea what that means?
el-tocino
on 24 Jul 2019
@el-tocino It doesn't seem to have something to do with wake-up.
hippyk
on 24 Jul 2019
Does it hear you correctly ever?
el-tocino
on 24 Jul 2019
The error message you're seeing is that the audio recorded doesn't contain any speech that the STT can transcribe.
This usually happen (as @el-tocino points out) when the listening is falsely triggered, i.e. the wake word spotter thinks it hears a "hey mycroft" from some piece of music or similar.
Other possibilities:
- If you trigger by hitting the IPC file and you don't speak directly the listening may timeout and miss your words.
- If there is some mic issue and the audio is missing or quality is very bad this could occur.
The last question by el-tocino is very relevant and I'm also curious to that.
forslund
on 24 Jul 2019
Does it hear you correctly ever?
Yes, occasionally succeed. And i use Google Voice search to test the microphone and it is perfectly ok.
@forslund
hippyk
on 25 Jul 2019
I just tried five more times, only one failed. It looks unstable.
hippyk
on 25 Jul 2019
Hi HippyK,
How are you triggering startListening and how often?
If this is not the case, please ignore, however if you are attempting the constant listening idea that you previously mentioned, our API will reject the requests. Using our server as a general STT service is not permitted. We provide this service for free to all users because we believe that everyone should have access to a privacy respecting voice assistant. Any attempts to abuse that service have detrimental impacts on the whole Community.
krisgesling
on 25 Jul 2019
Now it does look really because it didn't hear what I said. Maybe the first error tip can be optimized.
hippyk
on 25 Jul 2019
if i understand correctly you are manually creating the listening signal used by speech listener, why are you doing it this way? if you are doing it often (live transcription it seems to be the case) you will run into that all the time
1 - when audio is not understood, this would be most of the time if listening 24/7
2 - when requests are rate limited
if you are doing this please stop, that is not the intended use and hurts the whole community
you should look into streaming STT options, but be warned that is paid by the minute, you will also want to look into VAD detection to avoid sending "silence" to the STT and waste credits
something like Kaldi or deepspeech might be better suited for 24/7 listening if you have the resources to run those
JarbasAl
on 26 Jul 2019
Try and look into running this locally if you want to do non-stop listening: https://discourse.mozilla.org/t/new-project-deepspeech-websocket-server-client/32554
el-tocino
on 26 Jul 2019
@JarbasAI @el-tocino Thank you very much for your advice. But this problem doesn't have nothing to do with the way you wake up. Wake up by voice is the same. I just used the command line for convenience, and I didn't wake it up continuously.
hippyk
on 26 Jul 2019
Jarbas explains the error succinctly above. Without more complete details it will be difficult to provide relevant help.
el-tocino
on 26 Jul 2019
Related issues
krisgesling
路
5Comments
Ahmed-Benj
路
7Comments
tmajibon
路
9Comments
InconsolableCellist
路
8Comments
Zacki84
路
10Comments
Most helpful comment
if i understand correctly you are manually creating the listening signal used by speech listener, why are you doing it this way? if you are doing it often (live transcription it seems to be the case) you will run into that all the time
1 - when audio is not understood, this would be most of the time if listening 24/7
2 - when requests are rate limited
if you are doing this please stop, that is not the intended use and hurts the whole community
you should look into streaming STT options, but be warned that is paid by the minute, you will also want to look into VAD detection to avoid sending "silence" to the STT and waste credits
something like Kaldi or deepspeech might be better suited for 24/7 listening if you have the resources to run those