Rhasspy not picking up words correctly

[ghost]

So first of all, I'm using the Spanish profile since I'm Spanish, but I also tried the English profile and got the same results. It doesn't pick up the wake word, neither with porcupine in English or pocketsphinx in Spanish. Then if I manually wake Rhasspy up, it doesn't pick up what I say correctly neither. But really, not even close (in Spanish and English). Configuration is just as it comes. Just changed microphone input and sound output. I've really tried tweaking things but I can't seem to fix it. Microphone is not the best but it is decent enough. I really want to get this working so I'd appreciate any suggestion. I'm running Rhasspy on a raspberry pi 3b+.

[ghost]

So messing around I managed to get the wake word working, Spanish profile with porcupine selected as wake word. I really need to pronounce 'porcupine' in English to get it picked up. Still, transcriptions of commands are nowhere near what I actually say in Spanish.

[mathquis]

Porcupine only works for English phonemes so that’s not surprising...

Did you train you assistant with specific intents? Do you use open transcription? Are you using Pocketsphinx, Kaldi or Deepspeech as ASR?

[ghost]

Yes. I did not train it too deeply but just enough to test it. No open transcription. Pocketsphinx. Just changed to Pocketsphinx for wake word also, and it doesn't pick it up in Spanish.

[ghost]

Fixed the wake word. It appeared like an unknown word. But still, it doesn't pick right the command. Its nothing like what I actually say

[synesthesiam]

Thanks for giving this a try, @geezsanr. The Spanish profile is largely untested (outside of some Google TTS phrases), so I'd really like to work with you to make sure it's ok :)

It may be that the pocketsphinx model is poor, or the dictionary may use different spelling/casing/hyphenation than a real Spanish speaker would expect.

Can you post your sentences.ini? If you're willing too, a WAV file of you speaking a sentence that doesn't work would be very helpful.

Thanks!

[ghost]

Of couse @synesthesiam As I said my sentences.ini is not very large because I wanted to make sure it worked nice before going more in depth into it;

[GetTemperature]
qué tiempo hace
cuál es la temperatura

[GetTime]
qué hora es
cuál es la hora

[Music]
pon música
pausa la música

[SetAlarm]
pon una alarma

For now it has only picked up correctly one sentence. Wake word seems to work but it randomly wakes up a lot of times, and if I decrease the sensitivity it reaches a point where it doesn't pick it up anymore. I recorded myself speaking 6 sentences. I named the files after the sentence I'm saying but if you need specification or anything else just let me know :)

https://drive.google.com/file/d/14MLYF-GknzJ9q8VOuC56ntG5SEHN1CJB/view?usp=sharing

I hope the quality is good enough

[synesthesiam]

Thanks! Taking a look now...

[synesthesiam]

Hi @geezsanr, once I added "cuál es el tiempo" to [GetTemperature] all of the WAV files were successfully recognized.

Given the wake word issue you mentioned, I wonder if there is some background noise that Rhasspy is picking up?

Edit: Forgot to note that I'm also testing with Rhasspy 2.5! Can you give that a try?

[ghost]

Hi @geezsanr, once I added "cuál es el tiempo" to [GetTemperature] all of the WAV files were successfully recognized.

Given the wake word issue you mentioned, I wonder if there is some background noise that Rhasspy is picking up?

Edit: Forgot to note that I'm also testing with Rhasspy 2.5! Can you give that a try?

Oh really? I don't think there was any background noise, but I will give it a try again and see. I will post how it goes