Hi,

We've recently deployed a voice bot within the CAC1 region, yet have been experiencing significant difficulties with utterance recognition and background noise. We've tested the bot with noise sensitivity set to every step between 0 to 1, yet it seems to have no effect. Tests with wind noise, music, or speech (eg a tv) in the background tend to trigger the bot to immediately say "Sorry, I didn't get that" before the caller even has a chance to speak themselves.

Strangely though, we've imported the bot to a dev org on USE1 and it seems that the noise cancellation is significantly improved.

Has anyone else run into issues involving noise sensitivity, or also found a discrepancy in the sensitivity between regions?
Is there an "ideal" setting that the speech recognition expects, like a relatively quiet office environment instead of a noisy home?

Thanks,

Hello Shawn, 

I see you already have an active case opened for this issue. This would be my best recommendation as the sensitivity level should be making a difference and if its not you would need a Care engineer to review and escalate through the proper channels if necessary.

Hi,

We've recently deployed a voice bot within the CAC1 region, yet have been experiencing significant difficulties with utterance recognition and background noise. We've tested the bot with noise sensitivity set to every step between 0 to 1, yet it seems to have no effect. Tests with wind noise, music, or speech (eg a tv) in the background tend to trigger the bot to immediately say "Sorry, I didn't get that" before the caller even has a chance to speak themselves.

Strangely though, we've imported the bot to a dev org on USE1 and it seems that the noise cancellation is significantly improved.

Has anyone else run into issues involving noise sensitivity, or also found a discrepancy in the sensitivity between regions?
Is there an "ideal" setting that the speech recognition expects, like a relatively quiet office environment instead of a noisy home?

Thanks,

Hi,

We've recently deployed a voice bot within the CAC1 region, yet have been experiencing significant difficulties with utterance recognition and background noise. We've tested the bot with noise sensitivity set to every step between 0 to 1, yet it seems to have no effect. Tests with wind noise, music, or speech (eg a tv) in the background tend to trigger the bot to immediately say "Sorry, I didn't get that" before the caller even has a chance to speak themselves.

Strangely though, we've imported the bot to a dev org on USE1 and it seems that the noise cancellation is significantly improved.

Has anyone else run into issues involving noise sensitivity, or also found a discrepancy in the sensitivity between regions?
Is there an "ideal" setting that the speech recognition expects, like a relatively quiet office environment instead of a noisy home?

Thanks,

Hi Shawn,

Brad Carroll and I work together at AAA.  We have been struggling with our own recent experiences with voice bot speech recognition.  We have one flow in particular that handles tens of thousands of calls per day that appeared to be working very well until July 12 when we started noticing 2x-3x the number of bot "Ask for Yes/No" actions returning a NO value, however recordings and transcripts of this bot clearly show that the caller never said NO.  We just today turned off the barge-in setting as a workaround and speech reco appears to be functioning more accurately.   Another observation of ours is that the bot prompt playback was stopping much more quickly when barging in for calls before July 12 than it has since.  The bot continues playback for ~2 seconds after the caller barges in since July 12.  Accuracy with barge-in turned on also appears better with longer responses and much worse with very short (single, one syllable word) responses. 

Changing sensitivity level and other input or action related settings other than barge-in have had no perceptible impact on speech reco accuracy in our experience.

Your mileage may vary!  We have a ticket open with Genesys Cloud and are waiting for answers. 

All the best,

Frank Remski

Hi,

We've recently deployed a voice bot within the CAC1 region, yet have been experiencing significant difficulties with utterance recognition and background noise. We've tested the bot with noise sensitivity set to every step between 0 to 1, yet it seems to have no effect. Tests with wind noise, music, or speech (eg a tv) in the background tend to trigger the bot to immediately say "Sorry, I didn't get that" before the caller even has a chance to speak themselves.

Strangely though, we've imported the bot to a dev org on USE1 and it seems that the noise cancellation is significantly improved.

Has anyone else run into issues involving noise sensitivity, or also found a discrepancy in the sensitivity between regions?
Is there an "ideal" setting that the speech recognition expects, like a relatively quiet office environment instead of a noisy home?

Thanks,

We had a Genesys case in and got the following update over the weekend:

"We have received a response from our Dev team regarding your issue, and they have confirmed that it is related to an open Dev case that is being investigated. They have informed us that they have changed the main audio model we use for Google Speech-to-Text (STT) across all regions, which may have been causing the issue previously."

The functionality seems greatly improved now - we're able to tap on the mic and the bot will ignore the noise and give us a "I couldn't hear you" response instead.

Hi Shawn,

Brad Carroll and I work together at AAA.  We have been struggling with our own recent experiences with voice bot speech recognition.  We have one flow in particular that handles tens of thousands of calls per day that appeared to be working very well until July 12 when we started noticing 2x-3x the number of bot "Ask for Yes/No" actions returning a NO value, however recordings and transcripts of this bot clearly show that the caller never said NO.  We just today turned off the barge-in setting as a workaround and speech reco appears to be functioning more accurately.   Another observation of ours is that the bot prompt playback was stopping much more quickly when barging in for calls before July 12 than it has since.  The bot continues playback for ~2 seconds after the caller barges in since July 12.  Accuracy with barge-in turned on also appears better with longer responses and much worse with very short (single, one syllable word) responses. 

Changing sensitivity level and other input or action related settings other than barge-in have had no perceptible impact on speech reco accuracy in our experience.

Your mileage may vary!  We have a ticket open with Genesys Cloud and are waiting for answers. 

All the best,

Frank Remski

Hi,

We've recently deployed a voice bot within the CAC1 region, yet have been experiencing significant difficulties with utterance recognition and background noise. We've tested the bot with noise sensitivity set to every step between 0 to 1, yet it seems to have no effect. Tests with wind noise, music, or speech (eg a tv) in the background tend to trigger the bot to immediately say "Sorry, I didn't get that" before the caller even has a chance to speak themselves.

Strangely though, we've imported the bot to a dev org on USE1 and it seems that the noise cancellation is significantly improved.

Has anyone else run into issues involving noise sensitivity, or also found a discrepancy in the sensitivity between regions?
Is there an "ideal" setting that the speech recognition expects, like a relatively quiet office environment instead of a noisy home?

Thanks,